You are on page 1of 100

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/353499572

A Systematic Review of Traffic Conflict-Based Safety Measures with a Focus


on Application Context

Article  in  Analytic Methods in Accident Research · July 2021


DOI: 10.1016/j.amar.2021.100185

CITATIONS READS

0 23

5 authors, including:

Ashutosh Arun
CSIR-Central Road Research Institute
9 PUBLICATIONS   40 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Development of Indian Highway Capacity Manual View project

Surrogate Safety Evaluation of Signalised Intersections View project

All content following this page was uploaded by Ashutosh Arun on 29 July 2021.

The user has requested enhancement of the downloaded file.


1

1 A Systematic Review of Traffic Conflict-Based Safety Measures with a


2 Focus on Application Context

4 Ashutosh Arun
5 Doctoral Candidate, School of Civil and Environmental Engineering
6 Queensland University of Technology
7 Brisbane, Australia
8 E-mail: ashutosh.arun@hdr.qut.edu.au
9
10
11 Md. Mazharul Haque
12 (corresponding author)
13 Associate Professor, School of Civil and Environmental Engineering
14 Queensland University of Technology
15 Brisbane, Australia
16 E-mail: m1.haque@qut.edu.au
17
18
19 Simon Washington
20 Professor, School of Civil Engineering, Faculty of Engineering
21 University of Queensland
22 Brisbane, Australia
23 E-mail: simon@amagroup.io
24
25
26 Tarek Sayed
27 Professor, Department of Civil Engineering
28 University of British Columbia
29 Vancouver, Canada
30 E-mail: tsayed@civil.ubc.ca
31
32
33 Fred Mannering
34 Professor, College of Engineering
35 University of South Florida
36 Tampa, FL 33620, USA
37 E-mail: flm@usf.edu
38

39

40

41 May 2021
42
2

1 Abstract

2 Relative to safety assessment using data from observed crashes, conflict-based


3 road safety assessment can potentially provide additional insights into crash causation
4 processes. Despite numerous review studies on this topic, the application context of
5 conflict measures has been generally overlooked. This study conducts a systematic review
6 of conflict-based safety measures with a specific focus on the context of their
7 applications. This study employs the Preferred Reporting Items for Systematic Reviews
8 and Meta-Analyzes (PRISMA) guidelines of systematic review and meta-analysis to
9 review conflict measures used for the safety assessment of intersections over the past ten
10 years (2010-19). A total of 386 studies are systematically reviewed to identify conflict
11 measures used for various contexts, including intersection types, traffic operating
12 conditions, study types, and the purpose of the study.
13 The systematic review indicates that temporal proximity measures, specifically
14 time-to-collision and post-encroachment time, are the most widely used conflict measures
15 regardless of the application context. Other families of conflict measures such as spatial
16 proximity, kinematic, mixed and combinations of measures have also been applied
17 depending on the context. Using the extracted data from relevant studies, linear regression
18 models were developed for time-to-collision and post-encroachment time thresholds at
19 signalized intersections and time-to-collision thresholds at unsignalized intersections.
20 The thresholds are found to be associated with traffic environment types, sources of
21 conflict data and the application purpose of conflict measures. The findings of this study
22 identify several critical gaps in the literature that can help guide future research directions
23 in the conflict-based safety assessment of transport facilities. Critical gaps include the
24 scarcity of validation studies for conflict measures, the lack of suitable techniques to
25 estimate crash risk by severity types, the primary focus on signalized intersections
26 (leaving studies of other facility types underrepresented), and the lack of suitable conflict
27 measures for vulnerable road users.
28
29 Keywords: Conflict-based safety measures; traffic conflicts; conflict measure;
30 systematic review; threshold selection
31
3

1 1. INTRODUCTION

2 Conflict-based measures are generally used to delineate when a traffic event

3 evolves from a regular interaction into a traffic conflict and further into a crash. 1 One of

4 the most popular metrics is time-to-collision (Amundsen and Hyden, 1977). Time-to-

5 Collision (TTC) is defined as the time remaining to a potential collision if the interacting

6 road users' speed and direction remain unchanged (Hayward, 1972). Another popular

7 measure of temporal proximity is the Post-Encroachment Time (PET), which is defined

8 as the time between the departure of one road user and the arrival of another at a point or

9 an area of a potential collision (Allen et al., 1978). In addition to spatiotemporal proximity

10 measures, acceleration-based conflict measures have been applied. Previous review

11 studies have provided an excellent review of some of the conflict measures used in the

12 literature (Zheng et al., 2014b, Mahmud et al., 2017).

13 Nevertheless, most existing review studies are plagued by two crucial issues. First,

14 there are still several measures like the crash severity indicator Delta-V (Sobhani et al.,

15 2011) and other complex conflict measures such as Aggregated Crash Propensity Metric

16 (ACPM) (Wang and Stamatiadis, 2013) that have been generally overlooked in previous

17 review studies. Second, there is little to no discussion of the context in which the various

18 conflict measures of safety are applicable. The exclusion of context from the discussion

19 on suitable conflict measures can be problematic. For instance, Tageldin and Sayed

20 (2018) found that the suitability of conflict measures is dependent on the traffic

1
Conflict-based measures are also widely referred to as surrogate safety measures in much of the
extant literature. This terminology likely gives too much credit to actual crashes, essentially
acknowledging them as the correct measure of safety. In fact, the use of actual crash data creates
a model estimation problem in that drivers observed in crashes are not a representative sample
of the population since riskier drivers will be over-represented in crash data. This may make
identification of causal effects difficult (see Mannering et al. (2020) for a discussion of this
issue). Conflict-based measures do not completely resolve this issue since safe drivers may
avoid dangerous roadways, but they are likely far less influenced by this potentially serious data
problem.
4

1 environment, determined by the geographical location in which the traffic data are

2 collected. They studied pedestrian-vehicle conflicts in five cities worldwide, including

3 New York, New Delhi, Shanghai, Vancouver, and Doha, and found that temporal

4 proximity measures such as time-to-collision were useful conflict measures in the

5 organized traffic environments of New York and Vancouver. However, they were less

6 relevant as conflict measures in New Delhi and Shanghai's less-organized traffic

7 environments relative to kinematic measures such as evasive pedestrian maneuvers.

8 Similarly, other contextual variables, such as the study type and the application

9 purpose, can affect the choice of conflict measures and their thresholds. Johnsson et al.

10 (2018) reviewed several conflict measures for identifying conflicts involving vulnerable

11 road users and found that no universal indicator can satisfy all the various conditions

12 under different application contexts. Thus, they strongly argue in favor of choosing

13 conflict measures most relevant to the application context. Zheng et al. (2014b) also

14 observed that the lack of consensus on a standard set of conflict measures suggests that

15 appropriate measures may vary by application conditions.

16 Recently, Arun et al. (2021) conducted a systematic mapping review of the conflict-

17 based safety assessment and identified several topics critical to the comprehension and

18 advancement of conflict analysis. They flagged the non-consideration of the context while

19 selecting the conflict measures and their thresholds as a serious gap in the literature and

20 highlighted the lack of discussion in this regard in the current review studies available on

21 this topic (Kuang and Qu, 2014, Zheng et al., 2014, Mahmud et al., 2017). Similarly,

22 Zheng et al. (2021) reviewed the analytic methods in the modeling of traffic conflicts and

23 emphasized that the arbitrary selection of conflict measures and their thresholds had

24 hampered the cross-validation and generalization of the study findings.


5

1 Thus, following the research gaps highlighted by Arun et al. (2021) and Zheng et

2 al. (2021), this study reviews conflict measures of safety employed in the literature along

3 with the specific context of their application (application purpose, traffic environment,

4 study design, and so on). The objective of this study was to conduct a systematic review

5 of conflict-based safety measures to understand their application context

6 comprehensively and to identify critical research needs and future directions.

7 Specifically, this study seeks to do the following: 1) identify the various conflict measures

8 being used to measure safety at at-grade intersections; 2) identify the application context

9 of these various conflict measures; 3) determine how the study-specific context has

10 influenced the choice of conflict-based measures; 4) determine conflict measures that

11 have been used for identifying conflicts involving vulnerable road users, and 5) identify

12 the effect of context on conflict thresholds.

13
14 2. METHOD AND DATA

15 This study follows the Preferred Reporting Items for Systematic Reviews and

16 Meta-Analyzes (PRISMA) guidelines (Liberati et al., 2009) to conduct the systematic

17 search and review of the literature. The Population, Intervention, Comparator group,

18 Outcome, Study design Approach (PICOS) of formulating relevant and precise eligibility

19 criteria was followed in this study. The 'Population' of studies included all peer-reviewed

20 journals in the last ten years (2010-2019) dealing with traffic conflicts at at-grade

21 intersections, including signalized, stop-or-give way controlled, and roundabouts. The

22 'Intervention' ensured that only the studies about specific applications were considered.

23 They included conflict-based assessment of safety at intersections, development of new

24 conflict estimation methods, development of new conflict measures, establishing crash-

25 conflict relationships, and assessing the safety of vulnerable road users. Several studies

26 employing conflict-based methods also compared their safety results against results of
6

1 crash-based methods. Thus, the 'Comparator' group for conflict-based safety assessment

2 studies included studies conducted using traditional crash-based methods. The 'Outcome'

3 of conflict-based studies varied by intervention and were recorded accordingly.

4 Moreover, the 'Study design' for conflict-based studies included both original research

5 and case studies demonstrating the application of traffic conflicts in safety assessment.

6 The scientific databases used for searching the relevant literature included Scopus,

7 Web of Science, Compendex and Inspec (via Engineering Village), and ProQuest

8 databases. Further, a snowballing method (Wohlin et al., 2013) was adopted to identify

9 other relevant studies that could not be obtained from database searches. The following

10 search string yielded the best results on Scopus. Hence, it was followed on all the other

11 databases with minor tweaks and additional filters as required: (surrogate* OR conflict)

12 AND (road OR traffic OR crash) AND (intersection OR roundabout).

13 The search was limited to only the studies published within the ten-year period from

14 2010 to 2019 (inclusive) to focus on the most recent thinking in the field. The search was

15 further limited to peer-reviewed scholarly journal articles published in English to control

16 the quality and rigor of studies considered. These searches identified a total of 2070

17 records, including 579, 644, 659 and 187 articles in Scopus, Web of Science, Engineering

18 Village and ProQuest databases, respectively. Figure 1 shows the PRISMA flow diagram

19 of the number of studies included in every review stage. After removing the duplicate

20 records and papers unrelated to road transport and/or safety, the initial screening process

21 resulted in 527 studies. This initial set of papers was further screened following the

22 inclusion criteria of PICOS, resulting in 386 studies included in the qualitative synthesis

23 in this paper and 240 studies to be included in the quantitative synthesis in this paper.

24 In selecting articles for systematic review, it was essential to include studies that

25 have applied any conflict measures that can be defined by the following five specific
7

1 criteria, as proposed by Wu and Jovanis (2012): 1) conflict measures should have a short

2 observation period; 2) they should be correlated with a clinically meaningful outcome

3 (reduction in the frequency of conflict measures should reduce the frequency and/or

4 severity of crashes); 3) they should be statistically and causally related to crashes; 4) they

5 should completely capture the effect of a safety treatment in a way similar to how it would

6 affect crashes; and 5) they should be "markers" of crashes with a time scale, meaning that

7 they should be part of the same process as crashes, with crashes forming the endpoint of

8 that process.

9 The included 386 studies chosen for qualitative synthesis were further classified

10 using a systematic classification scheme shown in Figure 2. The first classification was

11 done to separate studies by the primary road user types, including motorized vehicles,

12 pedestrians, and cyclists. After that, the conflict-based safety studies involving motorized

13 vehicles were further classified into five classes based on the type of intersection facility,

14 including signalized intersections, stop-or-give way-controlled intersections,

15 roundabouts, J-turns/superstreets (alternatively called Reduced Conflict Intersections)

16 (Claros et al., 2017, Edara et al., 2015), and network. The network category included

17 studies that analyzed large road networks and did not give exact details about the types

18 and numbers of the various intersections in those networks (Dijkstra, 2013, Guo et al.,

19 2010b, Wu and Jovanis, 2013). For pedestrians and cyclists, because of the dearth of

20 studies on their conflicts at unsignalized intersections (both stop-and-give way-controlled

21 intersections and roundabouts), the conflict measures were not further classified. The 386

22 studies included for qualitative synthesis were critically reviewed to collect information

23 regarding the data items presented in Table 1.

24
8

1 3. RESULTS

2 The results of the systematic review according to the specific research questions of this

3 study are given below. The detailed definitions and formulae of all the conflict measures

4 reviewed in this study are given in Appendix A.

5
6 3.1 Conflict measures being used to measure safety at at-grade intersections

7 The systematic review identified a total of six families of conflict measures used

8 to measure safety at intersections, including temporal proximity measures, spatial

9 proximity measures, kinematic measures, mixed measures, a combination of measures,

10 and crash severity measures. These are summarized below.

11
12 3.1.1 Temporal proximity measures

13 Temporal proximity conflict measures include conflict indicators such as time-to-

14 collision and post-encroachment time as they measure the closeness of conflicting road

15 users to each other in time. Both time-to-collision and post-encroachment time were

16 found to be the most prevalent conflict measures irrespective of the context. Their

17 popularity may be attributed to the fact that, as Ismail et al. (2011) highlighted, both time-

18 to-collision and post-encroachment time are complementary indicators that measure

19 distinct aspects of a conflict. Time-to-collision calculates the proximity of the conflicting

20 road users to a collision point in the presence of a collision course (both the road users

21 arrive at the collision point simultaneously). On the other hand, post-encroachment time

22 measures the two road users' proximity to each other when one road user has left the

23 collision point. Although time-to-collision and post-encroachment time were generally

24 used for identifying all types of conflicts, most studies have adopted time-to-collision for

25 rear-end type conflicts. When road user trajectories have crossed each other, post-
9

1 encroachment times have been used as the additional suitable conflict measure (Mohamed

2 and Saunier, 2013, Mohamed and Saunier, 2015, Mohamed and Saunier, 2018).

3 Although the modified time-to-collision (Ozbay et al., 2008) was developed as an

4 improvement on the traditional time-to-collision by incorporating the vehicles'

5 acceleration information in its estimation and was found to have a higher correlation to

6 crashes than time-to-collision (Zheng and Sayed, 2019a), only four studies were found to

7 have used this measure for rear-end conflict identification. Other extensions of time-to-

8 collision, such as time-extended time-to-collision and time-integrated time-to-collision,

9 characterise a traffic conflict in terms of how long the time-to-collision value was below

10 its threshold and the extent to which the time-to-collision values fell below its threshold,

11 respectively, have similarly found fewer applications in conflict studies.

12

13 3.1.2 Spatial proximity measures

14 Several measures of spatial proximity, such as the proportion of stopping distance

15 and the potential index for collision with urgent deceleration, have been assessed by both

16 Zheng et al. (2014b) and Mahmud et al. (2017) in their review studies. However, only a

17 few studies have used them for the conflict-based safety assessment of intersections in

18 the past ten years. For instance, at signalized intersections, only one study (Oh et al.,

19 2010) used the stopping distance indicator for right turn-opposite direction and angle

20 conflicts. The reason for this low popularity may be attributed to the fact that spatial

21 proximity measures are more suited to measuring conflicts over longer distances, such as

22 those observed on a highway or arterial segments, as compared to the limited distances

23 available between road users in an intersection area.

24
10

1 3.1.3 Kinematic measures

2 The kinematic family of measures includes all the conflict measures that rely on

3 vehicle kinetic-state information such as speed and acceleration. Longitudinal

4 deceleration was found to be the most widely used kinematic measure. Deceleration was

5 chiefly utilized in identifying conflicts involving hard braking, defined as events where

6 the maximum deceleration exceeds a pre-set threshold (Lee et al., 2018, Songchitruksa

7 and Zha, 2014). Some studies also classify both acceleration and deceleration as a single

8 continuous acceleration indicator (Wei et al., 2019), where conflicts are identified based

9 on both the extreme positive (hard acceleration) and negative (hard braking) values of

10 this indicator. However, some studies (Lee et al., 2013, Pilko and Bared, 2010) only use

11 deceleration to measure the severity of the conflicts or their nearness to collisions. Other

12 specifically developed deceleration-based conflict measures, such as the minimum

13 required deceleration rate which a vehicle must apply to avoid a crash with the leading

14 vehicle, the crash potential index, and the brake threat number, have only seen limited

15 applications. Many studies have employed deceleration-based measures in addition to the

16 proximity measures for characterizing the conflicts (Essa and Sayed, 2019, So et al.,

17 2015a, So et al., 2015b, Zheng et al., 2019).

18

19 3.1.4 Mixed measures

20 Some conflict measures have comprised several conflict indicators in their

21 formulation and were classified as mixed measures in this study. For example, the

22 aggregated crash propensity metric (Wang and Stamatiadis, 2013) has been classified as

23 a mixed measure because it combines time-to-collision, reaction time, required

24 deceleration rate, and maximum available deceleration to measure safety. Among the

25 various mixed measures, near-crashes (based on the intensity of conflicts judged by


11

1 human experts) have been commonly used. The intensity of conflicts depends upon the

2 intensity of the evasive maneuvers, such as steering and acceleration/deceleration,

3 employed by one or both conflicting road users (Guo et al., 2010b). However, because it

4 is a measure based on subjective evaluation, there is no universally accepted

5 quantification of the intensity of evasive actions. Thus, various studies have employed

6 different parameters to identify a near-crash.

8 3.1.5 Combination of measures

9 If multiple measures were not combined into a single functional form, they were

10 labelled as a combination of measures. For example, So et al. (2015a) measured a conflict

11 event if the time-to-collision and the deceleration rate difference being measured

12 separately exceeded their respective thresholds simultaneously. Some other studies have

13 also employed various combinations of conflict measures to identify conflicts (Saunier et

14 al., 2011, Wang et al., 2019a, Ward et al., 2015, Zheng and Sayed, 2019d). The concept

15 behind such combinations is that individual conflict measures are partial images of the

16 complete safety picture and that using multiple conflict measures can capture all the safety

17 implications of a traffic interaction (Ismail et al., 2011). From the systematic review, it

18 was found that across all types of intersections, of the 386 unique studies included for the

19 qualitative synthesis (Figure 1), roughly 27% (106) used more than one conflict measure

20 for safety assessment. Again, time-to-collision and post-encroachment time were the most

21 commonly used conflict measures used in 46 (43%) of the 106 studies with multiple

22 measures. This observation is due to many simulation studies using the Surrogate Safety

23 Assessment Model (Gettman et al., 2008) for conflict identification, which provides

24 conflict output based on a standard set of six conflict measures, including time-to-

25 collision (threshold = 1.5s), post-encroachment time (threshold = 5s), the maximum speed
12

1 of either vehicle during the conflict, the difference between vehicular speeds at minimum

2 time-to-collision, initial deceleration of the subject vehicle, and maximum deceleration

3 of the subject vehicle. Out of these, most studies choose only to use the time-to-collision

4 and post-encroachment time measures. Also, recent studies have empirically

5 demonstrated that both time-to-collision and post-encroachment time in a bivariate

6 extreme value model setting can estimate crashes more accurately than either of them

7 alone or in combination with other conflict measures such as modified time-to-collision

8 and the deceleration rate to avoid a crash (Zheng and Sayed, 2019d, Zheng et al., 2019).

9 Zheng et al. (2019) demonstrated that both time-to-collision and post-encroachment time

10 are independent even at their extremes, meaning they capture different aspects of a

11 conflict event. While time-to-collision is an indicator of predicted danger in a conflict

12 event, post-encroachment time is the observed safety margin of the event once the

13 vehicles have crossed each other and the conflict has been resolved.

14 Other combinations of multiple conflict measures relate to the measurement of the

15 conflict severity (note that this conflict severity refers to the nearness to a collision, not

16 the injuries resulting from the collision). While a proximity measure is used to measure

17 the conflict, the nearness to a collision is measured by another conflict measure that

18 typically captures mechanical limitations of vehicles such as the required braking rate

19 (Babu and Vedagiri, 2017), driver capabilities in conflict situations such as reaction times

20 (Saunier et al., 2010), or both (Wang and Stamatiadis, 2013). A popular measure of

21 conflict severity is the severity index (Autey et al., 2012), a dimensionless measure based

22 on temporal proximity and perception-reaction time, typically used in combination with

23 time-to-collision or post-encroachment time. Swedish Traffic Conflict Technique

24 (Svensson and Hyden, 2006) measures conflict severity based on time-to-accident, which

25 is time to collision measured from the instant an evasive action was taken, and conflict
13

1 speed, which is the speed of the road user that takes the evasive action. This method

2 determines various levels of conflict severity and provides detailed charts based on the

3 ratio of time-to-accident to conflict speed to distinguish between the various levels of

4 conflict severity. Several studies (e.g., Sakshaug et al., 2010, Uzondu et al., 2018) have

5 successfully applied this method for disparate various types of intersections and traffic

6 environments. Longitudinal deceleration, as already mentioned before, has also been used

7 as a measure of conflict severity.

9 3.1.6 Crash severity measures

10 Another concept related to conflict-based safety assessment is the estimation of

11 crash severity using conflict measures. The expected collision severity is a function of the

12 kinetic energy expected to be released by each road user due to the collision. The collision

13 is assumed to be perfectly inelastic, with both the vehicles expected to stick and move

14 together post-collision for ease of calculation. However, only nine studies have employed

15 conflict measures for crash severity measurement, as summarized in Table 2.

16 A popular conflict measure for severity is the change in vehicle velocities resulting

17 from the collision (Delta-V). It was incorporated in the Surrogate Safety Assessment

18 Model (Gettman et al., 2008) software and has since been used in several studies. Since

19 the standard definition of Delta-V assumes constant velocity of conflict participants from

20 the time of the detection of conflicts, Laureshyn et al. (2017) introduced the concept of

21 extended Delta-V that modifies the final collision speed of the conflict participants based

22 on whether they undertake precautionary or emergency deceleration. Some conflict

23 measures, such as the aggregated severe crash metric (Wang and Stamatiadis, 2014a) and

24 the safety index (Alhajyaseen, 2015), encompass the estimation of both crash opportunity
14

1 and the resultant severity in a single metric, which is an interesting aspect that merits

2 further investigation.

4 3.2 Application context of conflict measures

5 The application context of a study is defined herein by four characteristics,

6 including the type of intersection, type of traffic environment, study type and the purpose

7 of the study (see Table 1). Among the intersection types, there is a distinction between

8 conflict measures chosen for signalized and unsignalized intersections. Further, among

9 unsignalized intersections, roundabouts eliminate all straight-line interactions between

10 the vehicles except on the approaches, requiring different conflict measures than those

11 used at stop-and-give way-controlled junctions.

12 As demonstrated by Tageldin et al. (2017a), the selection of conflict measures is

13 also dependent on the traffic environment in which the study is conducted. In this review,

14 two types of traffic environments are considered: organized traffic environments for

15 studies conducted in countries like the USA and UK, and less-organized traffic

16 environments for studies conducted in countries like China, India, and South Korea.

17 The selection of conflict measures was also found dependent on the type of study

18 (see Table 1). Five study types were observed: 1) field studies, where data were collected

19 from field observations under uncontrolled conditions using methods like video

20 recordings from roadside cameras or vehicle-mounted sensors in naturalistic studies; 2)

21 test-track studies, where the study team could manipulate the field conditions such as

22 signal timings; 3) driving simulator studies; 4) microsimulation studies, where legacy

23 traffic microsimulation software such as VISSIM, specialized applications such as

24 TRITONE (Giofre et al., 2013), and other methods such as cellular automata (Chai et al.,
15

1 2015) are used for conflict observations; and 5) theoretical studies that simply provided

2 the theoretical foundation for topics such as new conflict measures (Zhou et al., 2011).

3 Finally, the choice of conflict measures was found to depend on the overall purpose

4 of the study, as reported in Table 1. Note that in Table 1, under the “purpose of study”

5 category, the list (a) thru (m) may not be exhaustive (there can be other purposes, apart

6 from the ones observed in the last ten years, for which conflict measures can be applied).

7 Items (a) and (b) both refer to general safety assessment studies; item (b) focusses

8 specifically on those naturalistic studies that also recorded crashes during their

9 observation period, and hence provide deeper insights into crash mechanisms through

10 additional pre-crash human and vehicular behaviors not available in other study types.

11 Item (i) recorded studies using new methods of conflict identification, such as computer

12 vision-based automated methods (Saunier et al., 2010), or novel estimation models for

13 common conflict measures such as time-to-collision (Sobhani et al., 2013a), or new

14 methods of extracting conflict information from such conflict measures (Wei et al., 2019).

15 Novel technologies like connected and automated vehicles and safety-enhancing systems

16 like collision avoidance systems (items (j) thru (l)) require special conflict measures and

17 have been recorded separately. Finally, some studies focused only on the safety of specific

18 vehicle types such as buses or motorcycles that may require conflict measures distinct

19 from general-purpose safety assessment. The following section describes the effect of

20 these context variables on the selection of conflict measures.

21

22 3.3 Effects of study-specific context on the choice of conflict-based measures

23 3.3.1 Signalized intersections

24 Tables 3(a), 3(b) and 3(c) provide snapshots of various conflict measures used at

25 signalized intersections in the study period 2010-19, classified by temporal and spatial
16

1 proximity measures, kinematic measures, and mixed measures and combination of

2 measures. Both time-to-collision and post-encroachment time are the most prevalent

3 conflict measures at signalized intersections, with 84 studies (58%) reporting 100

4 applications of time-to-collision (applications for detecting different types of conflicts

5 counted separately) and 45 (31%) studies reporting 55 applications of post-encroachment

6 time for detecting various types of conflicts at signalized intersections. Among kinematic

7 measures, longitudinal deceleration was the most used with 16 applications from 13

8 studies.

10 3.3.2 Unsignalized intersections

11 Similar trends are also seen at unsignalized intersections in Tables 4(a) (temporal

12 proximity, spatial proximity and kinematic measures) and 4(b) (mixed measures and

13 combination of measures). With roundabouts (Table 5), temporal proximity measures,

14 particularly time-to-collision and post-encroachment time, have dominated conflict-

15 based safety assessment. In addition, since these intersections are characterized by

16 continuous traffic flow, gap-based indicators such as the critical gap (the time gap

17 between successive mainline vehicles) were also used by many studies.

18

19 3.3.3 Traffic environment

20 Regarding traffic environment, out of the 240 studies considered in the

21 quantitative synthesis (Figure 1), 149 (62%) were conducted in countries with organized

22 traffic flow environments such as the USA, Canada, and European countries, while the

23 rest in countries with less-organized traffic flow environment such as China, India, and

24 South Korea. Figures 3(a) and (b) give the conflict measures most used in both types of

25 environments. The measures represented in these figures have been used in more than
17

1 one study. Again, time-to-collision and post-encroachment time were the most used

2 conflict measures in both types of environments.

3 Notably, studies conducted in countries with organized traffic environments tend to

4 prefer temporal proximity measures, with more than 75% of studies employing measures

5 like gap and modified time-to-collision in addition to time-to-collision and post-

6 encroachment time. In contrast, studies in less-organized traffic environments have opted

7 for kinematic measures like longitudinal deceleration (7%) and critical speed (4%), as

8 well as mixed measures like near-crashes (10%). These results are as one might expect

9 because, as in organized traffic environments, vehicles tend to maintain safe headways

10 and, hence, any unusual interactions can be identified by detecting vehicles travelling

11 closer than usual. On the other hand, in less-organized traffic environments, vehicles tend

12 to move closer to each other in time and space; therefore, it is challenging to identify

13 conflicts based on proximity measures alone, and additional measures are required to

14 distinguish between normal and risky traffic situations (Tageldin et al., 2015). Moreover,

15 many studies in less-organized traffic environments rely on manual identification of

16 traffic conflicts from in-field or video observations as evidenced by higher usage of near-

17 crashes and the combination of time-to-accident and conflict speed as conflict measures;

18 the latter pioneered in the Swedish Traffic Conflict Technique (Hydén, 1987).

19

20 3.3.4 Type of study

21 Among the study types, most of the studies reviewed were either field (47%) or

22 microsimulation studies (44%). Only 5% of the conflict-based safety assessment studies

23 employed a driving simulator for traffic data collection. There was only a handful of test

24 track-based and theoretical studies (2% each). According to the type of the study, the

25 most used conflict measures are given in Figure 4 for field, microsimulation and driving
18

1 simulator categories. Figure 4 indicates that field studies have used a more comprehensive

2 range of conflict measures relative to microsimulation and driving simulator studies.

3 Among field studies, as expected, conflict measures used were heavily influenced

4 by the method of conflict observation. Figure 5 illustrates that computer vision-based

5 automated conflict analysis studies favored proximity measures like time-to-collision,

6 while naturalistic studies relying on in-vehicle instrumentation favored kinematic

7 measures like deceleration. Studies relying on loop detector data employed measures like

8 rear-end crash potential (Dimitriou et al., 2018) that primarily utilize headway and

9 occupancy information. Manual identification of conflicts is still quite prevalent; hence

10 evasive action-based measures like near-crashes that can be easily identified from field

11 observations are also popular in such studies. Lastly, a few studies also rely on public

12 databases such as the Safety Pilot Model Deployment that contains data collected from

13 field deployment of connected-and-automated vehicles and provides rich information

14 regarding vehicle kinematics such as speed, acceleration, and jerk profiles. To follow up

15 on this, some studies (Arvin et al., 2019, Kamrani et al., 2018, Wali et al., 2018) have

16 computed what they call driving volatility measures that capture the extreme variations

17 in vehicle kinematic parameters through statistical measures of dispersion such as

18 standard deviation and coefficient of variation.

19 Among simulation studies, 49 (53%) of the 93 microsimulation studies use the

20 Surrogate Safety Assessment Model (Gettman et al., 2008) with its standard conflict

21 measures like time-to-collision, post-encroachment time and deceleration. Consequently,

22 those three measures are prevalent and account for nearly 90% of all the conflict measure

23 applications in studies using microsimulation. In driving simulator-based studies, conflict

24 measures are typically used in two contexts; measures like time-to-collision are used for

25 defining and simulating conflict scenarios in the simulator (Yan et al., 2014, Yan et al.,
19

1 2015), while the safety outcome of driver reaction to conflicts can be studied by using

2 conflict measures such as post-encroachment time (Haque et al., 2016) and time-to-

3 intersection (Uchida et al., 2011).

5 3.3.5 Application purpose of the study

6 Given the many application purposes and the myriad of conflict measures used

7 for each of those purposes, only the top three conflict measures corresponding to each

8 purpose are presented in Figure 6. For conflict prediction functions, time-to-collision was

9 the most prevalent conflict measure. The conflict prediction models are like crash-based

10 count regression models and estimate conflicts as functions of intersections' geometric

11 and operational parameters (Sacchi and Sayed, 2016b, Sacchi and Sayed, 2016a).

12 Moreover, the development of gap-based conflict prediction functions was a dominant

13 study purpose for priority-based movements such as permitted right turns at signalized

14 intersections, merging from a minor road onto major road traffic at unsignalized

15 intersections, and entering the circulation at roundabouts (Dutta and Ahmed, 2018,

16 Pecchini et al., 2014, Zohdy et al., 2010).

17 Time-to-collision and post-encroachment time were again quite prevalent in studies

18 investigating the relationship between crashes and conflicts. Given that both time-to-

19 collision and post-encroachment time are zero at the time of the collision, they can be

20 readily used in extreme value theory models for crash prediction, whether individually,

21 as in univariate extreme value models (Wang et al., 2018, Zheng and Sayed, 2019c, Zheng

22 and Sayed, 2019b), or together in bivariate extreme value models. The latter models have

23 been found to increase the accuracy and precision of crash predictions in both organized

24 (Zheng et al., 2019) and less-organized (Wang et al., 2019a) traffic environments. Wang

25 et al. (2019a) found that the time-to-accident indicator is also a useful measure with time-
20

1 to-accident and time-to-collision in a bivariate model providing the best crash predictions

2 for rear-end conflicts. They did not evaluate the time-to-accident and time-to-collision

3 combination against the well-known time-to-collision and post-encroachment time

4 combination though, as they argued that the conflict properties measured by these metrics

5 were quite different. Specifically, they argued that both time-to-collision and post-

6 encroachment time capture the failure of the intended evasive action by a conflicting

7 vehicle, while time-to-accident, which is essentially time-to-collision at the start of the

8 evasive action, indicates the failure of perception-reaction on the part of the conflicting

9 road users. Driving volatility measures have also been used in crash prediction modeling,

10 although only as additional explanatory variables in traditional count data regression

11 models (Arvin et al., 2019).

12 Only one study (Scanlon et al., 2015) employed a naturalistic assessment of the

13 safety of intersections wherein they used longitudinal deceleration and yaw rate as

14 conflict measures; the latter defined as the rate of change of a vehicle's heading angle.

15 Given the simplicity of their estimation, time-to-collision and post-encroachment

16 time were the primary choices for studies that employed conflict measures for other

17 safety-related applications (Figure 6), such as driver behavior modeling (Pawar and Patil,

18 2018, Yan et al., 2018), development and assessment of new computer vision-based

19 models for conflict detection (Li et al., 2019, Saunier et al., 2010, Sayed et al., 2012,

20 Wang et al., 2019a, Xie et al., 2019, Zhang et al., 2019), multi-objective signal

21 optimization (Li and Sun, 2019, Stevanovic et al., 2013, Stevanovic et al., 2015), safety

22 assessment of vehicle- and infrastructure-based safety-enhancing ITS technologies (Li et

23 al., 2013, Songchitruksa and Zha, 2014, Tageldin et al., 2014, So et al., 2015a, Olia et al.,

24 2016, Morando et al., 2018, Virdi et al., 2019), and assessment of microsimulation models
21

1 (Saleem et al., 2014, Guo et al., 2019, Essa and Sayed, 2015b, Essa and Sayed, 2015a,

2 Essa and Sayed, 2016).

4 3.4. Conflict measures involving vulnerable road users

5 The results of the systematic review of conflict measures used for interactions

6 between vehicles and vulnerable road users and interactions among different vulnerable

7 road users themselves (pedestrians and cyclists) are given in Tables 6(a) and 6(b).

8 Analogous to the results of vehicle-vehicle conflicts, most of the studies have chosen to

9 use the typical proximity indicators; time-to-collision and post-encroachment time. Out

10 of the 88 unique studies reviewed, almost half of them used time-to-collision (42 studies,

11 48%), while a third of them used post-encroachment time (29 studies, 33%) for

12 identifying both vehicle-cyclists and vehicle-pedestrian conflicts. Note that some studies

13 use a variation of the post-encroachment time concept, variably called time difference to

14 the point of intersection (Wu et al., 2018) or time difference to collision (Zhang et al.,

15 2014), where the arrival times of the conflicting vehicle and pedestrian at the collision

16 point are predicted based on their trajectories, and the difference between them is used as

17 the conflict measure. Such measures are similar to the predicted post-encroachment time

18 measure used in some studies on vehicle conflicts (Mohamed and Saunier, 2017). Hence,

19 those studies have also been classified under post-encroachment time.

20 The method of conflict identification again plays a vital role, as highlighted by the

21 high use of evasive actions and near-crashes for manually identified conflicts in roughly

22 13% (20 studies) of the reviewed studies. Due to the rapidly dynamic nature of pedestrian

23 trajectories, such as instantaneous change of speed and directions, many studies dealing

24 with vulnerable road user conflicts have employed spatial proximity measures like the

25 distance between the stopping position of a vehicle and a pedestrian. Moreover, the
22

1 measurement of vehicle speeds is more reliable, and, in the case of a vehicle-vulnerable

2 road-user crash, the resultant injury to the vulnerable road user is almost entirely

3 dependent on the vehicle speed at the time of impact. Hence, some studies have used the

4 vehicle approach speed, typically in combination with some proximity measures like

5 time-to-accident or the distance between the vehicle and pedestrian (Wu et al., 2018), for

6 conflict identification. A similar pattern is observed in pedestrian-cyclist conflicts, where

7 some studies have utilized the spatial separation between the road users and the speed of

8 the larger road user (cyclists) for conflict identification.

9 Additionally, some studies have developed measures specifically for vulnerable

10 road user conflicts that typically analyze the volatility in the kinematic properties of the

11 vulnerable road users, such as the variations in speeds and accelerations, to identify abrupt

12 changes in their behavior that can be potential conflict indicators. Examples of such

13 measures include the maximum rate of change in step frequency of the pedestrians

14 (Tageldin and Sayed, 2016, Tageldin et al., 2017a) or permutation entropy of pedestrian

15 speed profile (Tageldin et al., 2017b).

16

17 3.5 The effect of context on conflict thresholds

18 The threshold for separating conflicts from regular traffic interactions is an

19 essential factor in conflict-based safety assessments. Tarko (2018) remarked that an

20 improper choice of thresholds could have severe implications for the validity of the crash-

21 conflict relationship. Moreover, Essa and Sayed (2015a) observed that the correlation

22 between simulated and field-observed conflicts is highly dependent on the threshold

23 adopted to identify conflicts.

24 As is the case with conflict measures, conflict threshold selection is also dependent

25 on the context. A simple regression model was estimated using thresholds selected from
23

1 the reviewed literature to get a historical sense of this relationship as a function of the

2 various contextual variables, as reported in Table 1. Table 7 presents the conflict

3 thresholds' descriptive statistics at signalized, unsignalized and roundabout intersections,

4 respectively. Conflict measures like near crashes do not require thresholds, and hence

5 they are not mentioned in this table. Also, not all studies provided information on the

6 threshold used. For example, Zhou et al. (2011) mentioned that they utilized the 85th

7 percentile values of time-to-collision as a conflict threshold but did not provide a value.

8 Hence, such studies could not be included in the analysis. Further, if studies used multiple

9 thresholds to identify conflicts, then either the maximum (or minimum) threshold used

10 was recorded, as in the case of Essa and Sayed (2018); or, if the study validated the safety

11 performance of the conflicts according to the various thresholds, then the threshold

12 yielding the best safety results were recorded. Also, for signalized intersections, only

13 time-to-collision and post-encroachment time have adequate sample sizes for regression

14 analysis, while, for unsignalized intersections, only time-to-collision has an adequate

15 sample size. Thus, only these conflict measures were utilized for further analysis. For

16 roundabouts, none of the conflict measures has an adequate sample size to enable

17 regression modeling. Thus, for other conflict measures like the deceleration rate to avoid

18 a crash, the threshold decisions can be made based on the median values provided in

19 Table 7.

20 Simple ordinary least squares regression models were estimated.2 Table 8 gives the

21 best fit models for time-to-collision and post-encroachment time thresholds for signalized

22 intersections and time-to-collision thresholds for unsignalized intersections. Given the

2
Random parameters models were also considered. However, the small sample properties of
ordinary least squares regression (with the relatively few observations available) made it the
preferred estimation mode relative to the simulated maximum likelihood estimation required for
random parameters model estimation (Washington et al., 2020).
24

1 low sample sizes, relatively few variables were significant in each of the models.

2 Interestingly, different context variables were significant for different conflict measures

3 at the two intersection types. For the time-to-collision threshold at signalized intersections

4 model, the indicator variables for organized traffic and simulation study were statistically

5 significant. Both variables reduced conflict thresholds used by about 1s.

6 For the post-encroachment time threshold at signalized intersections model, the

7 organized traffic indicator parameter increased the threshold used by a bit over 1.5s, and

8 the crash-conflict relationship indicator reduced the threshold used by nearly 1.7s.

9 Regarding the time-to-collision threshold for unsignalized intersections, only the

10 organized traffic indicator was statistically significant, resulting in nearly a 1.6s reduction

11 in the threshold used, meaning that studies in organized traffic environments have used

12 lower time-to-collision thresholds than in less organized traffic environments. This

13 finding is similar to that found in the signalized-intersection model (although there was

14 more than a 0.5s difference).

15 The traffic environment is clearly shown to be an essential factor in conflict

16 threshold selection, as it had a statistically significant effect in all three models presented

17 in Table 8. At signalized intersections, organized traffic environments have used a lower

18 time-to-collision threshold than studies conducted in less-organized traffic environments.

19 The effect of an organized traffic environment on time-to-collision threshold selection is

20 also negative for unsignalized intersections. This result seems counterintuitive given that

21 the latter's disorderly traffic movements should give rise to lower time-to-collisions than

22 the organized traffic environment, and, hence, studies should probably adopt lower

23 thresholds for detecting conflicts. It also stands in contrast to the post-encroachment time

24 model finding where the effect of the organized traffic environment variable generates a

25 positive parameter, indicating a significant number of studies in such traffic environments


25

1 have adopted a higher post-encroachment time threshold than studies in less-organized

2 traffic environments. A possible explanation for this could be related to the probabilistic

3 nature of time-to-collision estimation as opposed to the deterministic nature of post-

4 encroachment time. The orderly fashion of traffic movements in organized traffic

5 environments make it easier to measure speed and distance precisely, which is necessary

6 for time-to-collision calculation using automated methods like video cameras and

7 instrumented vehicles (Autey et al., 2012, Ward et al., 2015). On the other hand, conflict

8 studies in less-organized traffic environments still largely rely on manual conflict

9 identification methods and adopt higher time-to-collision thresholds to accommodate

10 human errors. Conversely, post-encroachment time measurement requires direct

11 observation of a single value, which is easier even with manual observation methods. It

12 is also possible that other factors, like the use of extreme value models for analyzing the

13 conflicts, may influence the adoption of time-to-collision thresholds in countries like

14 Canada (Zheng et al., 2014a). Indeed, the time-to-collision thresholds in Wang et al.

15 (2019a) that used extreme value models for conflict analysis in China are also lower than

16 the typically adopted threshold of 1.5 s in other studies like Tageldin et al. (2015) and

17 Chai and Wong (2014) conducted in similar traffic environments. This discussion

18 highlights the need for more studies investigating threshold selection specifically for

19 lesser-organized traffic environments.

20 This study has also found that the type of study and the application purpose were

21 significant determinants of conflict thresholds, indicating that future studies should be

22 extremely judicious in selecting conflict thresholds. The effect of the simulation-study

23 indicator at signalized intersections was negative, indicating that studies using this

24 method have adopted lower time-to-collision thresholds than studies adopting field and

25 driving simulator methods. The effect of the crash-conflict relationship indicator on post-
26

1 encroachment time threshold selection was negative, indicating that studies investigating

2 the relationship between the crashes and the conflicts typically adopt lower thresholds

3 than studies conducted for other purposes. The results regarding the effect of context

4 variables on threshold selection are also confirmed from the observations from other

5 studies. Songchitruksa and Zha (2014) noted that using the traditional time-to-collision

6 threshold of 1.5s produced eight times more conflicts than a threshold of 1.0s in the safety

7 assessment of connected vehicles, indicating that ignoring the study context can have a

8 deleterious effect on the validity of the study.

9 Unfortunately, the relative dearth of studies employing conflict measures other than

10 time-to-collision and post-encroachment time in various contexts meant that a similar

11 quantitative analysis could not be carried out for other conflict measures.

12

13 4. DISCUSSION AND FUTURE RESEARCH DIRECTIONS

14 This study has systematically reviewed the conflict measures used in the past ten

15 years (2010-19). Crucially, previous review studies on conflict measures have largely

16 overlooked the application context of conflict measures and their associated thresholds.

17 Thus, building on the work of Arun et al. (2021) and Zheng et al. (2021), which noted the

18 non-consideration of the context of conflict measures as a serious concern in the body of

19 knowledge, this study investigated the contexts of their application and provide insights

20 into how the context of a study influences the choice of conflict measures and their

21 thresholds. Towards this end, the study has summarized all the conflict measures used at

22 the various types of at-grade intersections, including signalized intersections (Tables 3(a)

23 to 3(c)), stop or give-way-controlled intersections (Tables 4(a) and 4(b)) and roundabouts

24 (Table 5) and discussed the most prevalent conflict measures according to contextual

25 variables such as traffic environment (Figure 3), study type (Figure 4 and 5), and study
27

1 purpose (Figure 6). This study has reviewed the conflict thresholds used in the literature

2 and summarized them in Table 7 for each intersection type. Further, the effect of context

3 on the selection of conflict threshold has been quantified through a simple ordinary least

4 squares regression (Table 8). This systematic review has identified critical research needs

5 and future research directions in this area that are briefly discussed below.

6
7 4.1 Definition of conflict measures

8 The most important topic of discussion arising from this review concerns the

9 definition of conflict measures and their distinction from the crash contributing factors.

10 As Arun et al. (2021) highlighted, conflicts need to be meaningfully related to crashes

11 and should be part of the same sequence of events that produce crashes. These criteria

12 place some constraints on what can be described as a traffic conflict measure.

13 Specifically, conflict measures need to be quantitative and continuously describe how

14 close a conflict event is to evolving into an observable crash event. This definition

15 precludes factors such as drunk driving, drowsiness, and violative actions such as

16 speeding because, while these factors contribute towards crash causation, they do not

17 clearly define a conflict event. Moreover, the various types of conflict measures reviewed

18 in this study can capture the effect of such factors. For instance, speeding in a car-

19 following situation will result in smaller time-to-collision values, and veering off the

20 roadway due to drowsiness or drinking can be captured by lateral acceleration and lateral

21 distance from the road edge. Moreover, there needs to be a vehicle in the cross-stream of

22 traffic for recording a possible crash in a violation involving red-light running.

23 Several studies have used traffic operational characteristics such as volume,

24 occupancy, and average speeds as conflict measures. These operational characteristics are

25 poor predictors of crash events because they do not have a meaningful relationship with

26 crashes. For instance, treatments to reduce traffic volume at an intersection may not
28

1 reduce crashes. Similarly, average speeds do not describe how a traffic interaction evolves

2 into a crash. Indeed, studies have used variations in speed profiles measured through 15th

3 and 85th percentiles and maximum speeds that, in a way, capture the effect of many

4 conflicts occurring at a location without measuring the conflicts themselves. Thus,

5 appropriate conflict measures need to be selected for conflict-based safety assessments.

7 4.2 Context-based selection of conflict measures

8 This review provides some evidence-based guidance for conflict measure and

9 threshold selection decisions for an array of contextual conditions. For instance, for

10 conducting a microsimulation-based site safety assessment of rear-end conflicts at

11 signalized intersections in an organized traffic environment, time-to-collision is the most

12 prevalent conflict measure.

13 This study has also underlined serious limitations in current knowledge concerning

14 conflict measures. Despite the development of many conflict measures over the years, the

15 earliest measures like time-to-collision and post-encroachment time are still the most

16 popular conflict measures (perhaps, because they are simple to measure and comprehend).

17 Moreover, their validity regarding measuring actual crash risk has been verified by many

18 studies under various contexts. However, this review distinctly demonstrates that a study's

19 context should be considered when choosing a suitable conflict measure. Zheng and

20 Sayed (2019a) compared four common conflict measures (time-to-collision, post-

21 encroachment time, modified time-to-collision and the deceleration rate to avoid a crash)

22 and found that the modified time-to-collision was the best performing indicator for rear-

23 end conflicts at signalized intersections. Although this study was informative, there is a

24 critical need for additional studies to compare conflict measures under various contexts

25 to identify the most suitable measure for each context.


29

1
2 4.3 Context-based selection of conflict thresholds

3 Similar to the context-based selection of conflict measures, this study provides

4 some insight into the context-based selection of thresholds. Notably, the simple

5 regression analysis for thresholds demonstrates that the context of a study influences

6 conflict-threshold selection. Of the variables found to be statistically significant in

7 threshold selection, the traffic environment is shown to be significant across all three

8 models presented in Table 8. In addition to the numerous points brought up in the earlier

9 discussion of the regression models shown in Table 8, most notably, the geographic

10 location related to organized and unorganized traffic environments is a consistently

11 significant factor in threshold selection. This observation underscores the importance of

12 geographic context and, more specifically, the driver behavior associated with different

13 geographic entities.3

14

15 4.4 Conflict studies on unsignalized intersections and roundabouts

16 This review has found a preponderance of conflict studies focusing on signalized

17 intersections only, and those concerning other intersection types, especially roundabouts,

18 are severely lacking. The fact that signalized intersections are typically more dangerous

19 than roundabouts may explain this lop-sided focus of research (Sadeq and Sayed, 2016,

20 St-Aubin et al., 2013). However, there is likely the need for a greater emphasis on

3
In addition to this geographic (spatial) element, there is also likely a temporal element (although
the number of observations available in this study did not allow us to explore this). That is,
driver behavior is likely changing over time particularly with the introduction of advanced safety
features on cars (lane departure warning, autonomous braking, etc.) changes in driver distraction
(smart phones, etc.), and so on. This implies that appropriate thresholds will likely change over
time, and that this change will continue to increase with the increasing introduction of advanced
safety features, the introduction of autonomous vehicles in the traffic stream, and other time-
varying factors. Please see Mannering (2018) for a general discussion of temporal instability in
the context of highway safety.
30

1 roundabouts, especially with the introduction of newer and unconventional roundabout

2 layouts such as turbo and elliptical is increasing (Vasconcelos et al., 2014, Giuffrè et al.,

3 2017, Tesoriere et al., 2018). Research is also lacking concerning the aspects of traffic

4 conflicts that can be captured by existing conflict measures at even different types of

5 conventional roundabouts like mini-roundabouts, double roundabouts, and roundabout

6 interchanges. Based on the prevailing traffic conditions and composition, the interactions

7 can vary during entering, circulating, and exiting maneuvers, which may necessitate

8 adjustments to the existing conflict measures or the development of entirely new

9 measures for such conflicts. In addition, a large proportion of crashes occur at

10 unsignalized (stop or give way-controlled) intersections (for example, more than half of

11 intersection crashes in Queensland, Australia, Queensland, 2019). Still, there is a

12 surprising lack of research on unsignalized intersection safety using traffic conflict

13 techniques, presenting a critical research need for the future.

14
15 4.6 Conflict studies in less-organized traffic environments

16 Most studies have undertaken conflict-based safety analysis in organized traffic

17 environments. Many such studies utilize the latest technological solutions like computer

18 vision-based automated tracking of road users. However, the studies conducted in less

19 organized traffic environments still predominantly employ manually identified near-

20 crashes. Since the manual methods rely on the judgment of the trained observers and are

21 prone to subjective bias, the validity and generalizability of such study results are

22 questionable. Previous research has shown that the usual conflict measures may not be

23 suitable for the traffic environments in developing countries, and there is a need to

24 develop custom conflict measures for these environments.

25

26 4.7 Specific measures for vulnerable road users-involved conflicts


31

1 Studies for vulnerable road users still use conflict measures like time-to-collision

2 that are primarily developed for capturing vehicular conflict mechanisms. Given the

3 dynamic mechanical capabilities of vulnerable road users (almost instantaneous changes

4 in speed, acceleration, and direction), the constant velocity assumption in measures like

5 the time-to-collision and the deceleration rate to avoid a crash may not be the most

6 appropriate for measuring conflicts involving vulnerable road users. The use of vehicle-

7 based conflict indicators for vulnerable road users can introduce significant conflict

8 identification errors that invalidate their relationship with actual crashes. In this regard,

9 some studies have shown promising results towards developing conflict measures that

10 specifically analyze the uncertainties in vulnerable road user movements, like the

11 permutation entropy of the speed profile or the change rate in pedestrian step frequency.

12 Such vulnerable road users-oriented measures can be combined with usual proximity

13 measures or vehicle-based kinematic measures, like deceleration and negative jerk, that

14 capture driver reaction to conflicts to enhance conflict measurement accuracy.

15 Crucially, recent advancements in vehicle technology and road design have thrown

16 up new challenges for safety researchers. The emergence of new modes of personal

17 mobility like e-scooters and e-bikes that can reach much higher speeds than even the

18 conventional bicycles flags the need to account for the variability in the types of

19 vulnerable road users while analyzing the patterns of their conflicts with motorized

20 vehicles. Moreover, novel road design concepts like shared spaces, where the vulnerable

21 road users and motorists legally occupy the same road space, can significantly affect the

22 nature of vulnerable road user-involved traffic conflicts. Thus, further research into

23 developing and validating vulnerable road users-specific conflict measures is a worthy

24 research direction.

25
32

1 4.8 Using multiple conflict measures

2 Individual conflict measures can only capture partial images of the whole safety

3 picture. Zheng et al. (2019) demonstrated that time-to-collision and post-encroachment

4 time provide better safety predictions when used together because they capture different

5 aspects of a conflict event. Furthermore, a crash becomes probable only when time-to-

6 collision is less than the road user's reaction time. Thus, proximity measures like time-to-

7 collision are indicative of the safe limits of human capabilities. Similarly, in the case of

8 the deceleration rate to avoid a crash, a crash is more probable if the deceleration rate to

9 avoid a crash is greater than the maximum available deceleration rate of the vehicle. Thus,

10 kinematic measures essentially capture the limit of vehicular capabilities.

11 Correspondingly, in theory, using proximity measures in combination with kinematic

12 measures may help increase the efficiency of conflict-based evaluations of traffic

13 facilities. Zheng et al. (2019) and Wang et al. (2019a) have investigated combinations of

14 measures like the deceleration rate to avoid a crash and maximum deceleration,

15 respectively, with time-to-collision and post-encroachment time in a bivariate extreme

16 value model. However, they did not find any improvement in the crash predictions,

17 indicating that there may not be any benefit in combining proximity and kinematic

18 measures. However, both studies had certain limitations, such as the limited size of

19 conflict data and improper thresholds of the kinematic measures. Moreover, studies like

20 Wang and Stamatiadis (2014b) have demonstrated the advantages of using proximity and

21 kinematic measures together in conflict analysis. Therefore, given the theoretical

22 attractiveness of the premise, it is argued that more such validation efforts with multiple

23 conflict measures, especially combining proximity and kinematic measures, should be

24 undertaken for various contexts in the future.

25
33

1 4.9 Measures of crash severity

2 Crash severity estimation through conflict measures of safety is a nascent field of

3 study, and not much work has been done in this regard. Despite the popularity of the post-

4 collision change in vehicle velocity (Delta-V) as a conflict measure in in-depth crash

5 investigation and crash reconstruction studies, their use in conflict-based safety

6 assessment has been limited. Research needs in this area pertain to improving the Delta-

7 V measure by relaxing its constant velocity assumption (Laureshyn et al., 2017).

8 Moreover, mixed measures such as the aggregated severe crash metric (Wang and

9 Stamatiadis, 2014a) and the safety index (Alhajyaseen, 2015) that simultaneously

10 measure crash opportunity as well as the resultant severity can offer a more

11 comprehensive view of safety at an intersection. However, no further research in this

12 direction has been conducted using either of these measures. Therefore, future research

13 into estimating the crash frequency and their severities through traffic conflicts is another

14 exciting research direction.

15

16 4.10 Inclusion of behavioral metrics into conflict measurement

17 There have been multiple efforts to include human behavioral cues in conflict

18 measurement.4 Studies such as Kuang et al. (2015) have argued for including

19 characteristics like perception-reaction time into the conflict measurement framework by

20 modifying prevalent conflict measures. Other relevant behavioral metrics such as risk-

21 taking (Tang et al., 2015) and emergency braking performance (Huertas-Leyva et al.,

4
Behavioral elements have long been used in the analysis of the injury severity levels observed
in actual crashes, where detailed behavioral data are available from police crash reports and
other sources (Savolainen et al., 2011; Bhat and Mannering, 2014). However, understandably,
accessing such behavioral data in a traffic conflict environment can be challenging. In the
absence of detailed behavioral data, it is possible to view the missing behavioral data as
unobserved heterogeneity in the model structure, and this is potentially a fruitful area for future
research as well (Mannering et al., 2016).
34

1 2019) can also potentially affect the outcome of a traffic conflict event. For instance,

2 Haque et al. (2016) found that the post-encroachment times of mobile phone-distracted

3 drivers were significantly shorter than those of undistracted ones during gap acceptance

4 at roundabouts, which increase their likelihood of crash involvements. Indeed, a recent

5 study found that the inclusion of driver behavioral characteristics such as tailgating as

6 covariates in location parameter estimation of a generalized extreme value model

7 improved the accuracy of crash predictions from conflicts (Cavadas et al., 2020). Thus,

8 the incorporation of behavioral cues into conflict estimation presents exciting future

9 research opportunities. Such research efforts could greatly benefit from the developments

10 in novel methods of conflict observation like naturalistic driving data collection and

11 driving simulator methods that provide the opportunity to capture many such behavioral

12 cues under various driving contexts.

13

14 4.11 Real-time conflict measurement

15 Crashes in peak hours on major urban arterials can cause a breakdown of traffic

16 flow, leading to major delays and a consequent spike in vehicular emissions, so there is a

17 strong incentive for managing safety in real-time because of the potentially far-reaching

18 consequences in improving overall traffic flow and emissions efficiency. Measuring real-

19 time traffic conflicts to avoid crashes can thus clearly aid authorities in better managing

20 their transport networks through proactive risk mitigation with actions such as real-time

21 signal optimization at signalized intersections. The availability of modern technologies

22 such as closed-circuit television cameras equipped with edge processing capabilities,

23 capable of performing low latency real-time risk estimations, and the imminent

24 introduction of connected and automated vehicles presents new opportunities for real-

25 time conflict measurement. In particular, the initial mixing of connected and automated
35

1 vehicles and traditional vehicles is likely to lead to a highly dynamic traffic environment

2 where human driver behavior is likely to be temporally unstable and in a continuous state

3 of evolution (Mannering, 2018). Assessing traffic conflicts in real-time could provide

4 valuable insights into this evolutionary behavior and serve as a basis to mitigate potential

5 adverse safety impacts. This assessment would be assisted by the fact that connected and

6 automated vehicles are typically equipped with several sensors such as video cameras,

7 global positioning systems, and laser-based proximity sensors, and they can thus serve as

8 probes to collect wide-ranging safety information and improve the existing conflict

9 measures. In addition, given the advancements in vehicle-to-everything communications,

10 connected and automated vehicles will soon be communicating safety-related signals to

11 other connected and automated vehicles and roadside devices like edge-enabled

12 surveillance cameras and smart traffic signals. This development can even lead to

13 evolving new and improved real-time conflict measures that incorporate corridor or

14 network-level real-time traffic and safety information and provide a totally new

15 perspective on the crash-conflict relationship. Some studies have initiated efforts in this

16 regard by analyzing the volatility of kinematic measures of connected vehicles at both

17 driver and intersection levels (Arvin et al., 2019, Kamrani et al., 2018, Wali et al., 2018),

18 thus developing new and improved conflict measures for real-time conflict measurement

19 presents an exciting research opportunity.

20

21 4.12 Improvements in road user representation

22 A critical concern in the current measurement of conflicts is inaccuracy in road

23 users' geometric representation. Video-based observation studies are especially afflicted

24 in this regard as they typically adopt a centroid (or an edge) representation of the road

25 users while calculating conflict measures that can introduce significant errors (Laureshyn
36

1 et al., 2010). Lately, some studies have attempted calculating conflict measures using a

2 three-dimensional box representation of the road users (Li et al., 2015); however, they

3 involve fitting predetermined three-dimensional boxes over two-dimensional detections

4 of vehicles, which may lead to accuracy issues. Thus, more research into an accurate

5 representation of road users for improved measurement of conflicts is required. One

6 possible research direction could be adopting a field theory that represents road users and

7 other obstacles in the road environment as repulsive potential fields (Mullakkal-Babu et

8 al., 2020).

10 5. CONCLUSION

11 The potential for conflict-based analyzes to improve safety has led to numerous

12 studies over the past decade that sought to identify traffic conflicts in a better manner

13 through ever-improved conflict measures. However, the glut of conflict measures in

14 literature has also created a problem because, in the wake of the lack of research regarding

15 the various types of conflict measures and the contexts for which they are suitable,

16 researchers have sometimes adopted conflict measures and their thresholds with scant

17 regard to their appropriateness. Arun et al. (2021) identified the absence of a systematic

18 review of conflict measures as a critical research gap, and the present study has been

19 geared towards addressing this gap by conducting a systematic review of conflict

20 literature from the last ten years following the PRISMA guidelines.

21 The study has provided a comprehensive summary of conflict measures according

22 to their families based on the characteristics measured, namely, temporal proximity,

23 spatial proximity, kinematic, mixed, and combination of measures. The study results have

24 firmly established that the selection of conflict measure depends not only on the type of

25 intersection facility being studied but also on other contextual variables such as the traffic
37

1 environment, the type of study, and the application purpose.

2 This study has flagged critical gaps in the literature that can guide future research

3 for conflict-based safety assessment. One identified critical research gap is the lack of

4 conflict-based studies for unsignalized intersections and roundabouts. There is a clear

5 need to develop and use context-specific conflict measures because the arbitrary usage of

6 conflict measures reduces their efficiency in capturing safety problems at specific traffic

7 locations and, in extreme cases, can lead to counterfactual results. This issue primarily

8 requires more validation studies for existing conflict measures, especially kinematic and

9 mixed measures, under various contexts to enable wider adoption. The use of context-

10 appropriate conflict measures and their thresholds can hold the key to providing conflicts-

11 based analysis that contributes significantly to our understanding of highway safety.

12 Ultimately, this study identifies numerous opportunities for future applications of traffic-

13 conflict analysis and improvements in traffic-conflict measurement. Perhaps most

14 importantly, the review in this study charts a direction for the role that traffic conflict

15 analyzes are likely to play in a future of rapidly changing driver behaviors and vehicle

16 technologies brought on by the continual improvement in vehicle safety features and the

17 future introduction of autonomous vehicles.

18
38

1
2 REFERENCES

3 Alhajyaseen, W., 2015. The integration of conflict probability and severity for the safety

4 assessment of intersections. Arabian Journal for Science and Engineering, 40 (2),

5 421-430.

6 Allen, B., Shin, B., Cooper, P., 1978. Analysis of traffic conflicts and collisions.

7 Transportation Research Record 667, 67-74.

8 Amundsen, F., Hyden, C., Proceedings: First Workshop on Traffic Conflicts. In:

9 Amundsen, F., Hyden, C., eds. First Workshop on Traffic Conflicts, 1977 Oslo,

10 Norway.

11 Archer, J., Young, W. 2010. The measurement and modellingmodeling of proximal safety

12 measures. Proceedings of the Institution of Civil Engineers - Transport, 163 (4),

13 191-201.

14 Arun, A., Haque, M., Bhaskar, A., Washington, S., Sayed, T., 2021. A systematic

15 mapping review of surrogate safety assessment using traffic conflict techniques.

16 Accident Analysis and Prevention, 153, 106016.

17 Arvin, R., Kamrani, M., Khattak, A. 2019. How instantaneous driving behavior

18 contributes to crashes at intersections: Extracting useful information from

19 connected vehicle message data. Accident Analysis and Prevention, 127, 118-133.

20 Astarita, V., Giofré, V., 2019. From traffic conflict simulation to traffic crash simulation:

21 Introducing traffic safety indicators based on the explicit simulation of potential

22 driver errors. Simulation ModellingModeling Practice and Theory, 94, 215-236.

23 Autey, J., Sayed, T., Zaki, M., 2012. Safety evaluation of right-turn smart channels using

24 automated traffic conflict analysis. Accident Analysis and Prevention, 45, 120-30.
39

1 Babu, S., Vedagiri, P., 2017. Traffic conflict analysis of unsignalisedsignalized

2 intersections under mixed traffic conditions. European Transport-Trasporti

3 Europei, 66, 10.

4 Baratian-Ghorghi, F., Zhou, H., Zech, W., 2016. Red-light running traffic violations: A

5 novel time-based method for determining a fine structure. Transportation

6 Research Part A, 93, 55-65.

7 Borenstein, M., Hedges, L., Higgins, J., Rothstein, H., 2011. Introduction to meta-

8 analysis, John Wiley & Sons, New York, NY.

9 Cavadas, J., Azevedo, C., Farah, H, Ferreira, A., 2020. Road safety of passing maneuvers:

10 A bivariate extreme value theory approach under non-stationary conditions.

11 Accident Analysis and Prevention, 134, 105315.

12 Chai, C., Wong, Y., 2014. Micro-simulation of vehicle conflicts involving right-turn

13 vehicles at signalized intersections based on cellular automata. Accident Analysis

14 and Prevention, 63, 94-103.

15 Chai, C., Wong, Y., Lum, K., 2015. Safety impacts of red light cameras at signalized

16 intersections based on cellular automata models. Traffic Injury Prevention, 16 (4),

17 374-379.

18 Chen, C., Liu, L., Qiu, T., Ren, Z., Hu, J., Ti, F., 2018. Driver's intention identification

19 and risk evaluation at intersections in the internet of vehicles. IEEE Internet of

20 Things Journal, 5 (3), 1575-1587.

21 Claros, B., Zhu, Z., Edara, P., Sun, C. 2017. Design guidance for J-turns on rural high-

22 speed expressways. Transportation Research Record 2618, 69-77.

23 Dijkstra, A., 2013. Assessing the safety of routes in a regional network. Transportation

24 Research Part C, 32, 103-115.


40

1 Dimitriou, L., Stylianou, K., Abdel-Aty, M., 2018. Assessing rear-end crash potential in

2 urban locations based on vehicle-by-vehicle interactions, geometric

3 characteristics and operational conditions. Accident Analysis and Prevention,

4 118, 221-235.

5 Dutta, M., Ahmed, M., 2018. Gap acceptance behavior of drivers at uncontrolled T-

6 intersections under mixed traffic conditions. Journal of Modern Transportation,

7 26 (2), 119-132.

8 Edara, P., Breslow, S., Sun, C., Claros, B., 2015. Empirical evaluation of J-turn

9 intersection performance. Transportation Research Record, 2486, 11-18.

10 Essa, M. and Sayed, T. 2019. Full Bayesian conflict-based models for real time safety

11 evaluation of signalized intersections. Accident Analysis and Prevention, 129,

12 367-381.

13 Essa, M., Sayed, T. ,2015a. Simulated traffic conflicts: Do they accurately represent field-

14 measured conflicts? Transportation Research Record 2514, 48-57.

15 Essa, M., Sayed, T., 2015b. Transferability of calibrated microsimulation model

16 parameters for safety assessment using simulated conflicts. Accident Analysis and

17 Prevention, 84, 41-53.

18 Essa, M., Sayed, T., 2016. A comparison between PARAMICS and VISSIM in estimating

19 automated field-measured traffic conflicts at signalized intersections. Journal of

20 Advanced Transportation, 50 (5), 897-917.

21 Essa, M., Sayed, T., 2018. Traffic conflict models to evaluate the safety of signalized

22 intersections at the cycle level. Transportation Research Part C, 89, 289-302.

23 Essa, M., Sayed, T., 2019. Full Bayesian conflict-based models for real time safety

24 evaluation of signalized intersections. Accident Analysis and Prevention, 129,

25 367-381.
41

1 Fu, Y., Li, C., Luan, T. H., Zhang, Y., Mao, G. 2018. Infrastructure-cooperative algorithm

2 for effective intersection collision avoidance. Transportation Research Part C, 89,

3 188-204.

4 Gelso, E., Sjöberg, J., 2017. Consistent threat assessment in rear-end near-crashes using

5 BTN and TTB metrics, road information and naturalistic traffic data. IEEE

6 Intelligent Transportation Systems Magazine, 9 (1), 74-89.

7 Gettman, D., Pu, L., Sayed, T. and Shelby, S. G. 2008. Surrogate safety assessment model

8 and validation. Final Report, FHWA-HRT-08-051, Federal Highway

9 Administration, United States Department of Transportation, Washington, DC.

10 Giofre, V., Maciejewski, M., Merkisz-Guranowska, A., Piątkowski, B., Astarita, V.,

11 2013. Real road network application of a new microsimulation tool: TRITONE.

12 Archives of Transport, 27-28 (3-4), 111-121.

13 Giuffrè, T., Trubia, S., Canale, A., Persaud, B., 2017. Using microsimulation to evaluate

14 safety and operational implications of newer roundabout layouts for European

15 Road networks. Sustainability (Switzerland), 9 (11), 2084.

16 Guido, G., Saccomanno, F., Vitale, A., Astarita, V., Festa, D. 2011. Comparing safety

17 performance measures obtained from video capture data. Journal of

18 Transportation Engineering, 137 (7), 481-491.

19 Guo, F., Klauer, S., Hankey, J., Dingus, T. 2010. Near crashes as crash surrogate for

20 naturalistic driving studies. Transportation Research Record, 2147, 66-74.

21 Guo, Y., Essa, M., Sayed, T., Haque, M., Washington, S., 2019. A comparison between

22 simulated and field-measured conflicts for safety assessment of signalized

23 intersections in Australia. Transportation Research Part C, 101, 96-110.


42

1 Haque, M., Oviedo-Trespalacios, O., Debnath, A., Washington, S., 2016. Gap acceptance

2 behavior of mobile phone-distracted drivers at roundabouts. Transportation

3 Research Record, 2602, 43-51.

4 Hayward, J., 1972. Near-miss determination through use of a scale of danger. Highway

5 Research Record, 384, 24-34.

6 Huang, F., Liu, P., Yu, H., Wang, W., 2013. Identifying if VISSIM simulation model and

7 SSAM provide reasonable estimates for field measured traffic conflicts at

8 signalized intersections. Accident Analysis and Prevention, 50, 1014-24.

9 Huertas-Leyva, P., Nugent, M., Savino, G., Pierini, M., Baldanzini, N., Rosalie, S., 2019.

10 Emergency braking performance of motorcycle riders: skill identification in a

11 real-life perception-action task designed for training purposes. Transportation

12 Research Part F, 63, 93-107.

13 Hurwitz, D., Knodler, M., Nyquist, B., Moore, D., Tuss, H., 2012. Evaluating the

14 potential of advanced vehicle detection systems in mitigating dilemma zone safety

15 conflicts. ITE Journal, 82 (3), 24-28.

16 Hutton, J., Bauer, K., Fees, C., Smiley, A., 2015. Evaluation of left-turn lane offset using

17 the naturalistic driving study data. Journal of Safety Research, 54, 5-15.

18 Hydén, C., 1987. The development of a method of traffic safety evaluations: The Swedish

19 Traffic Conflicts Technique. PhD, Lund University.

20 Ismail, K., Sayed, T., Saunier, N. 2011. Methodologies for aggregating indicators of

21 traffic conflict. Transportation Research Record, 2237, 10-19.

22 Johnsson, C., Laureshyn, A., De Ceunynck, T., 2018. In search of surrogate safety

23 indicators for vulnerable road users: a review of surrogate safety indicators.

24 Transport Reviews, 38 (6), 765-785.


43

1 Kamal, M., Imura, J., Hayakawa, T., Ohata, A., Aihara, K., 2015. A vehicle-intersection

2 coordination scheme for smooth flows of traffic without using traffic lights. IEEE

3 Transactions on Intelligent Transportation Systems, 16 (3), 1136-1147.

4 Kamrani, M., Arvin, R., Khattak, A., 2018. Extracting useful information from basic

5 safety message data: an empirical study of driving volatility measures and crash

6 frequency at intersections. Transportation Research Record, 2672, 290-301.

7 Kamrani, M., Wali, B., Khattak, A., 2017. Can data generated by connected vehicles

8 enhance safety? Transportation Research Record, 2659, 80-90.

9 Kuang, Y., Qu, X., 2014. A review of crash surrogate events. second international

10 conference on vulnerability and risk analysis and management (ICVRAM) and

11 the sixth international symposium on uncertainty, Modeling, and Analysis

12 (ISUMA). Liverpool, UK: American Society of Civil Engineers.

13 Kuang, Y., Qu, X., Weng, J., Etemad-Shahidi, A., 2015. How does the driver's perception

14 reaction time affect the performances of crash surrogate measures? PLoS One, 10

15 (9), e0138617.

16 Laureshyn, A., De Ceunynck, T., Karlsson, C., Svensson, A., Daniels, S., 2017. In search

17 of the severity dimension of traffic events: Extended Delta-V as a traffic conflict

18 indicator. Accident Analysis and Prevention, 98 (2017), 46-56.

19 Laureshyn, A., Svensson, A., Hyden, C. 2010. Evaluation of traffic safety, based on

20 micro-level behavioural data: theoretical framework and first implementation.

21 Accident Analysis and Prevention, 42 (6), 1637-46.

22 Lee, C., So, J., Ma, J., 2018. Evaluation of countermeasures for red light running by traffic

23 simulator-based surrogate safety measures. Traffic Injury Prevention, 19 (1), 1-8.


44

1 Lee, C., Yun, I., Choi, J., Ko, S., Kim, J., 2013. Evaluation of semi-actuated signals and

2 pedestrian push buttons using a microscopic traffic simulation model. KSCE

3 Journal of Civil Engineering, 17 (7), 1749-1760.

4 Li, J., Liu, Y., Tageldin, A., Zaki, M. H., Mori, G., Sayed, T., 2015. Automated region-

5 based vehicle conflict detection using computer vision techniques. Transportation

6 Research Record, 2528, 49-59.

7 Li, X., Sun, J., 2019. Turning-lane and signal optimization at intersections with multiple

8 objectives. Engineering Optimization, 51 (3), 484-502.

9 Li, Y., Lu, J., Xu, K., 2017. Crash risk prediction model of lane-change behavior on

10 approaching intersections. Discrete Dynamics in Nature and Society 2017,

11 7328562.

12 Li, Z., Wei, H., Yao, Z., Xiong, H., Yan, X., 2013. Optimized advance detector

13 configuration for option zone protection at high-speed signalized intersections.

14 Transportation Research Record, 2355, 60-72.

15 Liberati, A., Altman, D., Tetzlaff, J., Mulrow, C., Gotzsche, P., Ioannidis, J., Clarke, M.,

16 Devereaux, P., Kleijnen, J., Moher, D., 2009. The PRISMA statement for

17 reporting systematic reviews and meta-analyseanalyzes of studies that evaluate

18 health care interventions: explanation and elaboration. PLoS Med, 6 (7),

19 e1000100.

20 Liu, J., Cai, B., Wang, Y., Wang, J., 2013. A lane level positioning-based cooperative

21 vehicle conflict resolution algorithm for unsignalized intersection collisions.

22 Computers and Electrical Engineering, 39 (5), 1381-1398.

23 Liu, J., Khattak, A. 2016. Delivering improved alerts, warnings, and control assistance

24 using basic safety messages transmitted between connected vehicles.

25 Transportation Research Part C, 68, 83-100.


45

1 Ma, Y., Qin, X., Grembek, O., Chen, Z. 2018. Developing a safety heatmap of

2 uncontrolled intersections using both conflict probability and severity. Accident

3 Analysis and Prevention, 113, 303-316.

4 Mahmud, S., Ferreira, L., Hoque, M., Tavassoli, A., 2017. Application of proximal

5 surrogate indicators for safety evaluation: A review of recent developments and

6 research needs. IATSS Research, 41 (4), 153-163.

7 Mannering, F., 2018. Temporal instability and the analysis of highway accident data.

8 Analytic Methods in Accident Research 17, 1-13.

9 Mannering, F., Bhat, C., 2014. Analytic methods in accident research: Methodological

10 frontier and future directions. Analytic Methods in Accident Research 1, 1-22.

11 Mannering, F., Bhat, C., Shankar, V., Abdel-Aty, M., 2020. Big data, traditional data and

12 the tradeoffs between prediction and causality in highway-safety analysis.

13 Analytic Methods in Accident Research 25, 100113.

14 Mannering, F., Shankar, V., Bhat, C., 2016. Unobserved heterogeneity and the statistical

15 analysis of highway accident data. Analytic Methods in Accident Research 11, 1-

16 16.

17 Mohamed, M., Saunier, N. 2013. Motion prediction methods for surrogate safety analysis.

18 Transportation Research Record, 2386, 168-178.

19 Mohamed, M., Saunier, N., 2015. Behavior analysis using a multilevel motion pattern

20 learning framework. Transportation Research Record, 2528, 116-127.

21 Mohamed, M., Saunier, N., 2018. The impact of motion prediction methods on surrogate

22 safety analysis: A case study of left-turn and opposite-direction interactions at a

23 signalized intersection in Montreal. Journal of Transportation Safety and Security,

24 10 (4), 265-287.
46

1 Morando, M., Tian, Q., Truong, L., Vu, H., 2018. Studying the safety impact of

2 autonomous vehicles using simulation-based surrogate safety measures. Journal

3 of Advanced Transportation, 2018.

4 Mullakkal-Babu, F. A., Wang, M., He, X., van Arem, B., Happee, R. 2020. Probabilistic

5 field approach for motorway driving risk assessment. Transportation Research

6 Part C, 118, 102716.

7 Muralidharan, A., Coogan, S., Flores, C., Varaiya, P. 2016. Management of intersections

8 with multi-modal high-resolution data. Transportation Research Part C, 68, 101-

9 112.

10 Noble, A., Dingus, T., Doerzaph, Z., 2016. Influence of in-vehicle adaptive stop display

11 on driving behavior and safety. IEEE Transactions on Intelligent Transportation

12 Systems, 17 (10), 2767-2776.

13 Oh, J., Kim, E., Kim, M., Choo, S. 2010. Development of conflict techniques for left-turn

14 and cross-traffic at protected left-turn signalized intersections. Safety Science, 48

15 (4), 460-468.

16 Olia, A., Abdelgawad, H., Abdulhai, B., Razavi, S., 2016. Assessing the potential impacts

17 of connected vehicles: mobility, environmental, and safety perspectives. Journal

18 of Intelligent Transportation Systems, 20 (3), 229-243.

19 Ozbay, K., Yang, H., Bartin, B., Mudigonda, S. 2008. Derivation and validation of new

20 simulation-based surrogate safety measure. Transportation Research Record,

21 2083, 105-113.

22 Paul, M., Ghosh, I. 2018. Speed-based proximal indicator for right-turn crashes at

23 unsignalized intersections in India. Journal of Transportation Engineering Part A:

24 Systems, 144 (6).


47

1 Pawar, D., Patil, G., 2017. Minor-street vehicle dilemma while maneuvering at

2 unsignalized intersections. Journal of Transportation Engineering Part A:

3 Systems, 143 (8).

4 Pawar, D., Patil, G., 2018. Response of major road drivers to aggressive maneuvering of

5 the minor road drivers at unsignalized intersections: A driving simulator study.

6 Transportation Research Part F, 52, 164-175.

7 Pecchini, D., Mauro, R., Giuliani, F. 2014. Model of potential crash rates of rural

8 roundabouts with geometrical features. Journal of Transportation Engineering,

9 140 (11).

10 Peesapati, L., Hunter, M., Rodgers, M., 2013. Evaluation of Postencroachment Time as

11 Surrogate for Opposing Left-Turn Crashes. Transportation Research Record,

12 2386, 42-51.

13 Peng, H., 2018. Study on collaborative control of vehicle system at road intersection in

14 dynamic competitive environment. Romanian Journal of Information Science and

15 Technology, 21 (2), 172-180.

16 Pilko, P., Bared, J., 2010. Safety of intersection phase change interval. ITE Journal, 80

17 (4), 42-48.

18 Preuk, K., Stemmler, E., Schiessl, C., Jipp, M., 2016. Does assisted driving behavior lead

19 to safety-critical encounters with unequipped vehicles' drivers? Accident Analysis

20 and Prevention, 95 (Pt A), 149-56.

21 Queensland Government. 2019. Crash data from Queensland roads [Online]. Available:

22 https://www.data.qld.gov.au/dataset/crash-data-from-queensland-roads

23 [Accessed Access 2019].


48

1 Rahman, M. S., Abdel-Aty, M., Lee, J., Rahman, M., 2019. Safety benefits of arterials’

2 crash risk under connected and automated vehicles. Transportation Research Part

3 C, 100, 354-371.

4 Sacchi, E., Sayed, T., 2016a. Conflict-based safety performance functions for predicting

5 traffic collisions by type. Transportation Research Record, 2583, 50-55.

6 Sacchi, E., Sayed, T., 2016b. Bayesian estimation of conflict-based safety performance

7 functions. Journal of Transportation Safety and Security, 8 (3), 266-279.

8 Sadeq, H., Sayed, T., 2016. Automated roundabout safety analysis: diagnosis and remedy

9 of safety problems. Journal of Transportation Engineering, 142 (12).

10 Sakshaug, L., Laureshyn, A., Svensson, A., Hyden, C., 2010. Cyclists in roundabouts--

11 different design solutions. Accident Analysis and Prevention, 42, 1338-1351.

12 Saleem, T., Persaud, B., Shalaby, A., Ariza, A., 2014. Can microsimulation be used to

13 estimate intersection safety? Transportation Research Record, 2432 (1), 142-148.

14 Sander, U., Lubbe, N,. 2018. Market penetration of intersection AEB: Characterizing

15 avoided and residual straight crossing path accidents. Accident Analysis and

16 Prevention, 115, 178-188.

17 Santiago-Chaparro, K., Qin, X., Noyce, D., 2010. Proposed safety index based on risk-

18 taking behavior of drivers. Transportation Research Record, 2147, 51-57.

19 Saunier, N., Mourji, N., Agard, B. 2011. Mining microscopic data of vehicle conflicts

20 and collisions to investigate collision factors. Transportation Research Record,

21 2237, 41-50.

22 Saunier, N., Sayed, T., Ismail, K. 2010. Large-scale automated analysis of vehicle

23 interactions and collisions. Transportation Research Record 2147, 42-50.


49

1 Savolainen, P., Mannering, F., Lord, D., Quddus, M., 2011. The statistical analysis of

2 crash-injury severities: A review and assessment of methodological alternatives.

3 Accident Analysis and Prevention 43(5), 1666-1676.

4 Sayed, T., Ismail, K., Zaki, M., Autey, J., 2012. Feasibility of computer vision-based

5 safety evaluations: Case study of a signalized right-turn safety treatment.

6 Transportation Research Record, 2280, 18-27.

7 Scanlon, J., Kusano, K., Gabler, H., 2015. Analysis of driver evasive maneuvering prior

8 to intersection crashes using event data recorders. Traffic Injury Prevention, 16

9 Suppl 2, S182-189.

10 Sharma, A., Bullock, D., Peeta, S. 2011. Estimating dilemma zone hazard function at high

11 speed isolated intersection. Transportation Research Part C, 19 (3), 400-412.

12 Shen, J., Tian, Z., Wang, W., 2013. Correlation between degree of complexity and

13 efficiency for intersections in China. Journal of Transportation Engineering, 139

14 (2), 193-199.

15 So, J., Dedes, G., Park, B., HosseinyAlamdary, S., Grejner-Brzezinsk, D. 2015a.

16 Development and evaluation of an enhanced surrogate safety assessment

17 framework. Transportation Research Part C, 50, 51-67.

18 So, J., Lim, I., Kweon, Y., 2015b. Exploring traffic conflict-based surrogate approach for

19 safety assessment of highway facilities. Transportation Research Record, 2513,

20 56-62.

21 Sobhani, A., Young, W., Bahrololoom, S., Sarvi, M. 2013a. Calculating time-to-collision

22 for analysing right turning behaviour at signalisedsignalized intersections. Road

23 and Transport Research, 22 (3), 49-61.


50

1 Sobhani, A., Young, W., Logan, D., Bahrololoom, S., 2011. A kinetic energy model of

2 two-vehicle crash injury severity. Accident Analysis and Prevention, 43 (3), 741-

3 54.

4 Sobhani, A., Young, W., Sarvi, M. 2013b. A simulation based approach to assess the

5 safety performance of road locations. Transportation Research Part C, 32, 144-

6 158.

7 Songchitruksa, P., Zha, L. 2014. Advancing safety performance monitoring at signalized

8 intersections through use of connected vehicle technology. Transportation

9 Research Record, 2432, 99-109.

10 St-Aubin, P., Saunier, N., Miranda-Moreno, L., Ismail, K., 2013. Use of computer vision

11 data for detailed driver behavior analysis and trajectory interpretation at

12 roundabouts. Transportation Research Record, 2389, 65-77.

13 Stevanovic, A., Stevanovic, J., Kergaye, C. 2013. Optimization of traffic signal timings

14 based on surrogate measures of safety. Transportation Research Part C, 32, 159-

15 178.

16 Stevanovic, A., Stevanovic, J., So, J., Ostojic, M., 2015. Multi-criteria optimization of

17 traffic signals: Mobility, safety, and environment. Transportation Research Part

18 C, 55, 46-68.

19 Svensson, A., Hyden, C., 2006. Estimating the severity of safety related behaviour.

20 Accident Analysis and Prevention, 38, 379-385.

21 Tageldin, A., Sayed, T. 2018. Models to evaluate the severity of pedestrian-vehicle

22 conflicts in five cities. Transportmetrica A: Transport Science, 1-22.

23 Tageldin, A., Sayed, T., 2016. Developing evasive action-based indicators for identifying

24 pedestrian conflicts in less organized traffic environments. Journal of Advanced

25 Transportation, 50 (6), 1193-1208.


51

1 Tageldin, A., Sayed, T., Shaaban, K. 2017a. Comparison of time-proximity and evasive

2 action conflict measures: case studies from five cities. Transportation Research

3 Record, 2661, 19-29.

4 Tageldin, A., Sayed, T., Wang, X. 2015. Can time proximity measures be used as safety

5 indicators in all driving cultures? Case study of motorcycle safety in china.

6 Transportation Research Record 2520, 165-174.

7 Tageldin, A., Sayed, T., Zaki, M., Azab, M., 2014. A safety evaluation of an Adaptive

8 Traffic Signal Control system using Computer Vision. Advances in

9 Transportation Studies, 2 (SPECIAL ISSUE), 83-96.

10 Tageldin, A., Zaki, M., Sayed, T., 2017b. Examining pedestrian evasive actions as a

11 potential indicator for traffic conflicts. IET Intelligent Transport Systems, 11 (5),

12 282-289.

13 Tang, K., Di, D., Li, K.. 2015. Risk-taking behavior of left-turners in gap acceptance and

14 its effects on capacity estimation at signalized intersections. Transportation

15 Research Record, 2483, 1-9.

16 Tarko, A., 2018. Estimating the expected number of crashes with traffic conflicts and the

17 Lomax Distribution - A theoretical and numerical exploration. Accident Analysis

18 and Prevention, 113, 63-73.

19 Tesoriere, G., Campisi, T., Canale, A., Zgrablić, T., 2018. The surrogate safety appraisal

20 of the unconventional elliptical and turbo roundabouts. Journal of Advanced

21 Transportation, 2018, 2952074.

22 Uchida, N., de Waard, D., Brookhuis, K., 2011. Countermeasures to prevent detection

23 failure of a vehicle approaching on collision course. Applied Ergonomics, 42 (4),

24 540-547.
52

1 Ulak, M, Ozguven, E., Moses, R., Sando, T., Boot, W., AbdelRazig, Y., Sobanjo, J., 2019.

2 Assessment of traffic performance measures and safety based on driver age and

3 experience: A microsimulation based analysis for an unsignalized T-intersection.

4 Journal of Traffic and Transportation Engineering (English Edition), 6 (5), 455-

5 469.

6 Uzondu, C., Jamson, S., Lai, F., 2018. Exploratory study involving observation of traffic

7 behaviour and conflicts in Nigeria using the Traffic Conflict Technique. Safety

8 Science, 110, 273-284.

9 Vasconcelos, L., Silva, A., Seco, A., Fernandes, P., Coelho, M., 2014. Turboroundabouts:

10 multicriterion assessment of intersection capacity, safety, and emissions.

11 Transportation Research Record, 2402, 28-37.

12 Virdi, N., Grzybowska, H., Waller, S., Dixit, V., 2019. A safety assessment of mixed

13 fleets with connected and autonomous vehicles using the surrogate safety

14 assessment module. Accident Analysis and Prevention, 131, 95-111.

15 Wali, B., Khattak, A., Bozdogan, H., Kamrani, M., 2018. How is driving volatility related

16 to intersection safety? A Bayesian heterogeneity-based analysis of instrumented

17 vehicles data. Transportation Research Part C, 92, 504-524.

18 Wang, C., Stamatiadis, N., 2013. Surrogate safety measure for simulation-based conflict

19 study. Transportation Research Record, 2386, 72-80.

20 Wang, C., Stamatiadis, N., 2014a. Derivation of a new surrogate measure of crash

21 severity. Transportation Research Record, 2432, 37-45.

22 Wang, C., Stamatiadis, N., 2014b. Evaluation of a simulation-based surrogate safety

23 metric. Accident Analysis and Prevention, 71, 82-92.


53

1 Wang, C., Xu, C., Dai, Y. 2019a. A crash prediction method based on bivariate extreme

2 value theory and video-based vehicle trajectory data. Accident Analysis and

3 Prevention, 123, 365-373.

4 Wang, C., Xu, C., Xia, J., Qian, Z., Lu, L. 2018. A combined use of microscopic traffic

5 simulation and extreme value methods for traffic safety evaluation. Transportation

6 Research Part C, 90, 281-291.

7 Wang, F., Tang, K., Li, K., Liu, Z., Zhu, L. 2019b. A group-based signal timing

8 optimization model considering safety for signalized intersections with mixed

9 traffic flows. Journal of Advanced Transportation, 2019, 2747569.

10 Wang, X., Khattak, A., Liu, J., Masghati-Amoli, G., Son, S. 2015. What is the level of

11 volatility in instantaneous driving decisions? Transportation Research Part C, 58,

12 413-427.

13 Ward, J., Agamennoni, G., Worrall, S., Bender, A., Nebot, E., 2015. Extending time to

14 collision for probabilistic reasoning in general traffic scenarios. Transportation

15 Research Part C, 51, 66-82.

16 Washington, S., Karlaftis, M., Mannering, F., Anastasopoulos, P., 2020. Statistical and

17 econometric methods for transportation data analysis. Third Edition, CRC Press,

18 Taylor and Francis Group, New York, NY.

19 Wei, Y., Li, K., Tang, K. 2019. Trajectory-based identification of critical instantaneous

20 decision events at mixed-flow signalized intersections. Accident Analysis and

21 Prevention, 123, 324-335.

22 Wohlin, C., Runeson, P., da Mota Silveira Neto, P., Engström, E., do Carmo Machado,

23 I., de Almeida, E., 2013. On the reliability of mapping studies in software

24 engineering. Journal of Systems and Software, 86 (10), 2594-2610.


54

1 Wu, J., Xu, H., Zheng, Y., Tian, Z. 2018. A novel method of vehicle-pedestrian near-

2 crash identification with roadside LiDAR data. Accident Analysis and Prevention,

3 121, 238-249.

4 Wu, K., Jovanis, P., 2012. Crashes and crash-surrogate events: exploratory modeling with

5 naturalistic driving data. Accident Analysis and Prevention, 45, 507-16.

6 Wu, K.-F., Jovanis, P., 2013. Screening naturalistic driving study data for safety-critical

7 events. Transportation Research Record, 2386, 137-146.

8 Xiao, Y., Ran, Q., Yang, J., Wang, Z., 2011. Analysis and modeling of crossing behavior

9 at urban intersections in China. Journal of Transportation Engineering 137 (2),

10 121-127.

11 Xie, K., Ozbay, K., Yang, H., Li, C., 2019. Mining automatically extracted vehicle

12 trajectory data for proactive safety analytics. Transportation Research Part C, 106,

13 61-72.

14 Yan, X., Xue, Q., Ma, L., Xu, Y., 2014. Driving-simulator-based test on the effectiveness

15 of auditory red-light running vehicle warning system based on time-to-collision

16 sensor. Sensors (Switzerland), 14 (2), 3631-3651.

17 Yan, X., Zhang, X., Xue, Q., 2018. How does intersection field of view influence driving

18 safety in an emergent situation? Accident Analysis and Prevention, 119, 162-175.

19 Yan, X., Zhang, Y., Ma, L. 2015. The influence of in-vehicle speech warning timing on

20 drivers’ collision avoidance performance at signalized intersections.

21 Transportation Research Part C, 51, 231-242.

22 Zhang, Q., Yang, Z., Sun, D. 2019. Automated data collection and safety analysis at

23 intersections based on a novel video processing system. Transportation Research

24 Record, 2673 (4), 136-144.


55

1 Zhang, Y., Yao, D., Peng, L., Qiu, T., 2014. Scene-based pedestrian safety performance

2 model in mixed traffic situation. IET Intelligent Transport Systems, 8 (3), 209-

3 218.

4 Zheng, L., Ismail, K., Meng, X., 2014a. Freeway safety estimation using extreme value

5 theory approaches: a comparative study. Accident Analysis and Prevention, 62,

6 32-41.

7 Zheng, L., Ismail, K., Meng, X., 2014b. Traffic conflict techniques for road safety

8 analysis: open questions and some insights. Canadian Journal of Civil

9 Engineering, 41 (7), 633-641.

10 Zheng, L., Sayed, T., 2019a. Comparison of traffic conflict indicators for crash estimation

11 using peak over threshold approach. Transportation Research Record 2673 (5),

12 493-502.

13 Zheng, L., Sayed, T., 2019b. Application of extreme value theory for before-after road

14 safety analysis. Transportation Research Record, 2673 (4), 1001-1010.

15 Zheng, L., Sayed, T., 2019c. A full Bayes approach for traffic conflict-based before-after

16 safety evaluation using extreme value theory. Accident Analysis and Prevention,

17 131, 308-315.

18 Zheng, L., Sayed, T., 2019d. From univariate to bivariate extreme value models:

19 Approaches to integrate traffic conflict indicators for crash estimation.

20 Transportation Research Part C, 103, 211-225.

21 Zheng, L., Sayed, T., Essa, M. 2019. Validating the bivariate extreme value modeling

22 approach for road safety estimation with different traffic conflict indicators.

23 Accident Analysis and Prevention, 123, 314-323.


56

1 Zheng, L., Sayed, T., Mannering, F., 2021. Modeling traffic conflicts for use in road

2 safety analysis: A review of analytic methods and future directions. Analytic

3 Methods in Accident Research, 29, 100142.

4 Zhou, S., Sun, J., An, X., Li, K., 2011. The development of a conflict hazardous

5 assessment model for evaluating urban intersection safety. Transport, 26 (2), 216-

6 223.

7 Zohdy, I., Sadek, S., Hesham, R., 2010. Empirical analysis of effects of wait time and

8 rain intensity on driver left-turn gap acceptance behavior. Transportation

9 Research Record 2173, 1-10.

10

11
57

Figure 1. Preferred Reporting Items for Systematic Reviews and Meta-Analyzes


(PRISMA) flow diagram for the systematic review
58

Figure 2. Systematic classification scheme formulated in the study


59

SI Delta- DVM MTTC TA, Jerk SI


3% V Gap 2% 2% Conf. S CS 2% 2%
3% 3% 3% 4%
DRAC
4% Decn
NC 7% TTC
4% TTC Gap 40%
Decn 49% 7%
5%
NC
10%
PET
25% PET
25%
(a) (b)

Legend
Abbreviation Full form Abbreviation Full form
CS Critical Speed Gap Gap Time
Conf. S Conflict Speed Jerk Negative Vehicular Jerk
Decn Longitudinal Deceleration MTTC Modified Time-to-Collision
Delta-V Change in vehicle velocity due to collision NC Near-crash
DRAC Deceleration Rate to Avoid a Crash PET Post-encroachment Time
DVM Driving Volatility Measures SI Severity Index

Figure 3. Most used conflict measures in countries with (a) an organized traffic
environment and (b) a less-organized traffic environment
60

80
70
Driving Simulator Field Simulation

60
Number of Studies

50
40
30
20
10
0

Conflict Measures
Legend
Abbreviation Full Form Abbreviation Full Form
ACPM Aggregated Crash Propensity Metric Gap Gap Time
Coll. Prob. Collision Probability NC Near-crash
Conf. S Conflict Speed PET Post-encroachment Time
CPI Crash Potential Index TA Time-to-Accident
Delta-S Relative Speed of Conflicting Vehicles TIT Time-Integrated Time-to-Collision
DRAC Deceleration Rate to Avoid a Crash TTC Time-to-Collision
DVM Driving Volatility Measures YR Yaw Rate

Figure 4. Most used conflict measures according to the three prominent types of study,
namely, driving simulator, field, and simulation studies
61

Number of Studies
0 10 20 30

Computer Vision TTC


Conlict Observation Methods

In-Vehicle Instrumentation Decn


Loop Detectors RECP
Manual Field Observations NC
Manual Identification from Video NC
Multiple Sensors TTI
NGSIM TTC
SPMD DVM

Legend
Abbreviation Full Form Abbreviation Full Form
Decn Longitudinal Deceleration RECP Rear-End Crash Potential
DVM Driving Volatility Measures SPMD Safety Pilot Model Deployment
NC Near-crash TTC Time-to-Collision
NGSIM Next Generation Simulation TTI Time-to-Intersection/Stop Line

Figure 5. Most used conflict measures according to the methods of conflict observation
in field studies
62

40
35
30
Number of Studies

25
20
15
10
5
0

ACPM
TTC

TTC

DVM
TTC

TTC
SI

TTC

TTC

TTC

TTI
TTC

TTC

TTC
YR
PET

DRAC
PET

Decn

PET

PET

PET

PET

PET

PET

PET

PET
Saf. I

Decn

NC

Decn
RI
Gap
Gap

CPI
CPF CCR DBM NSA NM NSSM SAwRU SO SMA SITS SSA V+SITS VITS
Application Purpose of the Study
Legend
Abbreviation Full-Form Abbreviation Full-Form Abbreviation Full-Form
ACPM Aggregated Crash Propensity Metric Gap Gap Time SI Severity Index
CCR Crash-Conflict Relationship NC Near-crash SITS Site-based Intelligent Transport System
CPF Conflict Prediction Function NM New Model SO Signal Optimisation
CPI Crash Potential Index NSA Naturalistic Safety Assessment SMA Simulation Model Assessment
DBM Driver Behaviour Modeling NSSM New Surrogate Safety Measure SSA Site Safety Assessment
Decn Longitudinal Deceleration PET Post-encroachment Time TTC Time-to-Collision
DRAC Deceleration Rate to Avoid a Crash Saf. I Safety Index VITS Vehicle-based Intelligent Transport System
DVM Driving Volatility Measures SAwRU Safety Assessment wrt Specific Road User V+SITS Vehicle + Site-based Intelligent Transport System
Figure 6. Most used conflict measures according to the application purpose of the study
63
64

Table 1. Data collected in this review study

Category Data Item


Study identifier • Author/s
• Year of publication
Type of intersection • Signalized
• Unsignalized (Stop-or-Give Way controlled)
• Roundabout
• J-turns
• Network
Type of conflict • All types
• Rear-end
• Angle
• Right-turn opposite direction (RTOD)*
• Sideswipe/lane change
• Head-on
• Single vehicle conflict
Conflict measure/s The safety conflict measure/s (SSM) used in the study; a binary
identifier [0,1] indicated if a single or multiple conflict
measures were used
Family of conflict • Temporal proximity measure
measure • Spatial proximity measure
• Kinematic measure
• Mixed measure
• Combination of measures
Crash dimension • Nearness to collision
captured by the • Severity of the crash outcome
SSM • Both nearness to collision and severity
• None (additional measures computed in a study to
characterise a conflict)
SSM threshold The threshold used in the study to identify conflicts; a binary
identifier [0,1] indicated if the threshold was based on
anecdotal evidence/previous studies or empirically estimated
Geographical The country/region where the data collection for the study was
region conducted
Type of study • Field study
• Test track study
• Driving simulator study
• Microsimulation study; a binary identifier [0,1] indicated if
the study used standard Surrogate Safety Assessment
Model (SSAM) thresholds
• Theoretical study
65

Category Data Item


Purpose of study The application purpose of the study:
a) Site safety assessment
b) Naturalistic safety assessment
c) Estimation of a new conflict measure
d) Estimation of crash-conflict relationship
e) Estimation of conflict prediction models
f) Modeling of safety-related driver behaviour
g) Signal optimisation (for signalized intersections)
h) Assessment of traffic conflict-specific microsimulation
models
i) Assessment of new models/methods of conflict
identification from field data
j) Assessment of vehicle-based intelligent transport system
application (advanced driver assist, collision
warning/avoidance system etc.)
k) Assessment of facility-based intelligent transport system
application (advanced traffic signal control)
l) Assessment of combined vehicle- and facility-based
intelligent transport system application
m) Safety assessment with respect to a specific type of vehicle
(such as motorcyclists or buses)
Method of The method of observation of conflicts (only for field studies):
observation • Manual field observations
• Manual identification of conflicts from videos
• Computer vision-based conflict identification from video
• In-vehicle instrumentation
• Data from another study
• Unspecified
66

Table 2. Conflict measures used for crash severity estimation (see Appendix B for a definition of terms)

Safety Conflict
Study Definition Formula
Measure
Aggregated Wang and ASCM is the average of the probabilities of If RT < TTC; or
Severe Crash Stamatiadis (2014a) occurrence of a crash of a given severity-type If RT>TTC AND
𝑣2 −𝑣1
Metric (ASCM) that is estimated based on time-to-collision 𝐷𝑅𝐴𝐶 = 2(TTC−RT) > 𝑀𝐴𝐷𝑅; then
(TTC), maximum available deceleration rate
𝑆𝑒𝑣𝑒𝑟𝑒 𝐶𝑟𝑎𝑠ℎ 𝑀𝑒𝑡𝑟𝑖𝑐 (𝑆𝐶𝑀)
(MADR) and driver's reaction time (RT), while
the severity probability is calculated as the ∆𝑣1 𝛽 ∆𝑣2 𝛽
= ( ) + ( )
power model of delta-V (∆𝑣). α α
∆𝑣1 𝛽 ∆𝑣2 𝛽
− ( ) ( )
α α
𝐴𝑆𝐶𝑀 = ∑ 𝑆𝐶𝑀
∀𝑖,𝑗
where, 𝑖, 𝑗 are conflicting pairs of vehicles
Collision Energy Astarita and Giofré The product of a distracted driver's kinetic 𝐿 1
𝐶𝐸 = . 2. 𝑚1 (𝑣1 sin 𝜃)2 . 𝑝
(CE) (2019) energy and the proportion of the energy that 𝑣1 𝛿𝑡 2
would be transferred to the subject vehicle due where, 𝐿 = length of the road segment (m), 𝛿𝑡 =
to the deviation from a straight trajectory simulation time step (s),
during the distraction duration. and p = proportion of energy relative to the
energy generated by a straight trajectory crash
Conflict Alhajyaseen (2015), The released kinetic energy of a collision 𝛼∆𝐾𝐸
𝑆𝑎𝑓. 𝐼 = 𝛽𝑃𝐸𝑇
Index/Safety Wang et al. (2019b) weighted by the probability of occurrence of 𝑒
Index (Saf. I) the collision based on post-encroachment time
(PET).
Delta-V Sobhani et al. Change in the velocity of a vehicle because of 𝑚2
∆𝑣1 = √𝑣 2 + 𝑣2 2 − 2𝑣1 𝑣2 cos 𝜃
(2013b), Laureshyn a collision if both the vehicles continue to 𝑚1 + 𝑚2 1
et al. (2017), Sander move at a constant velocity from the time of 𝑚1
∆𝑣2 = √𝑣 2 + 𝑣2 2 − 2𝑣1 𝑣2 cos 𝜃
𝑚1 + 𝑚2 1
67

Safety Conflict
Study Definition Formula
Measure
and Lubbe (2018), detection of conflict until the assumed
Ulak et al. (2019) collision.

Extended Delta-V Laureshyn et al. Change in the vehicle's velocity because of a 𝑣𝑜1 − 𝑎1 𝑡, 𝑖𝑓 (𝑣𝑜1 − 𝑎1 𝑡) ≥ 0
𝑣1 = {
(Ext. Delta-V) (2017) collision if the vehicles brake at a constant 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
deceleration rate from conflict detection until 𝑣𝑜2 − 𝑎2 𝑡, 𝑖𝑓 (𝑣𝑜2 − 𝑎2 𝑡) ≥ 0
𝑣2 = {
the assumed collision. 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
where, 𝑣𝑜1 and 𝑣𝑜2 are the initial speeds (m/s) of
the subject and conflicting vehicles, respectively,
at the start of the conflict,
t = duration of conflict (s).
Then ∆𝑣 is calculated as the formula given above
Injury Severity Sobhani et al. The expected occupant injury severity of a 𝐼𝑆𝑆 = ∑ (𝛼0 + 𝛼𝑖 𝑥𝑖 )
Score (ISS) (2013b) conflict, measured using the estimated kinetic ∀𝑖
energy of the subject vehicle (KE) and the where, 𝑥𝑖 are the independent variables (KE and
impact type of the expected crash, i.e., near- the type of impact)
side, far-side, front, or rear impact
Kinetic Energy Ma et al. (2018) The kinetic energy released because of a 1 1
∆𝐾𝐸 = (𝑣1 2 + 𝑣2 2 ) − 𝑣1 𝑣2 cos 𝜃
Loss per Unit collision between vehicles of comparable 4 2
Mass (∆KE) masses.

Notations
𝑣1 Speed of the subject vehicle (m/s) 𝑚1 Mass of the subject vehicle (kg) 𝑎1 Acceleration of the subject vehicle (m/s2)
𝑣2 Speed of the conflicting vehicle (m/s) 𝑚2 Mass of the conflicting vehicle (kg) 𝑎2 Acceleration of the conflicting vehicle (m/s2)
𝜃 Angle of deflection from the planned trajectory (due to distraction or collision) α, β Estimable parameters
68

Table 3(a). Conflict measures used at signalized intersections (2010-19): Temporal and spatial proximity measures
Spatial
Temporal Proximity Measures Proximity
Measures
Conflict Type BRT ET Gap MTTC PET PL SI TET TIT TTB TTC TTI SD
All types 1 22 4 41
Angle 2 1 8 1 11 1 1
Rear-end 1 4 9 3 1 1 1 28 2
RTOD 1 4 10 1 1 16 1
Sideswipe/Lane Changing 6 4
Total 1 3 6 4 55 1 9 1 1 1 100 3 2

Abbreviation Full Form Abbreviation Full Form Abbreviation Full Form


BRT Brake Reaction Time PET Post-encroachment Time TET Time-Extended Time to Collision
ET Encroachment Time PL Pre-emptive Level TIT Time-Integrated Time to Collision
Gap Gap Time SI Severity Index TTB Time to Brake
MTTC Modified Time to Collision SD Stopping Distance TTC Time to Collision
TTI Time to Intersection/Stop-line

Table 3(b). Conflict measures used at signalized intersections (2010-19): Kinematic measures
Kinematic Measures
Conflict Type Accn BTN CPI Curv. Decn Delta-V DRAC DVM Ext. Delta-V Jerk Lat A RA RS YR

All types 1 1 1 6 1 2 4 2 1 1 2
Angle 2 1
Rear-end 1 3 2
RTOD 5 2 1 2 1
Sideswipe/Lane Changing
Total 1 1 1 1 16 3 4 4 1 2 2 1 1 4

Abbreviation Full-Form Abbreviation Full-Form Abbreviation Full-Form


Accn Longitudinal Acceleration Decn Longitudinal Deceleration Lat A Lateral Acceleration
BTN Brake Threat Number DRAC Deceleration Rate to Avoid a Crash RA Required Longitudinal Acceleration
69

Abbreviation Full-Form Abbreviation Full-Form Abbreviation Full-Form


CPI Crash Potential Index DVM Driving Volatility Measures RS Relative Speed/Speed Difference
Curv. Change of Curvature Ext. Delta-V Extended Delta-V YR Yaw Rate

Table 3(c). Conflict measures used at signalized intersections (2010-19): Mixed measures and combination of measures
Mixed Measures Combination of Measures
Conflict Type

TTC, DRD
EA, Delta-

TTC, PET
V, Conf. S

TA, TTC

TTC, LR
TA, PET
P(UEA)
CHAM

MTTC

MTTC
ACPM

ASCM

DRAC

DRAC
RECP
LDSI

Saf. I

TTC,

TTC,
ICRI

PET,

PET,
DCP

NC
CE

CI
All types 3 1 1 1 5 1 2 1
Angle 1 1 2 1 1 1
Rear-end 1 1 1 1 1 1 1 1
RTOD 1 1 4 2 1
Sideswipe/Lane Changing 1
Total 3 1 1 1 2 2 1 1 11 1 1 3 1 1 1 1 1 1 2 1 1 2

Abbreviation Full-Form Abbreviation Full-Form Abbreviation Full-Form


ACPM Aggregated Crash Propensity Metric Conf. S Conflict Speed LR Loom Rate
ASCM Aggregated Severe Crash Metric DCP Dilemma Zone Conflict Potential NC Near Crash
CE Collision Energy DRD Deceleration Rate Difference P(UEA) Probability of Unsuccessful Evasive
Action
CHAM Conflict Hazardous Assessment Model ICRI Integrated Conflict Risk Index RECP Rear-end Conflict Potential
CI Conflict Intensity LDSI Left-turn Driver Safety Index Saf. I Safety Index
70

Table 4(a). Conflict measures used at unsignalized intersections (2010-19): Temporal proximity, spatial proximity, and kinematic measures
Spatial
Temporal Proximity Measures Proximity Kinematic Measures
Measures
Conflict Type Gap PET TA TTC TTC-TTAvoid TTI DTI CS Decn RS Delta- DRAC DVM Lat A RDR YR
V
All types 11 3 21 1 1 1 4
Angle 5 8 1 9 1 1 1 3 3 1 1 1 1
Rear-end 1 3 4 1
RTOD 1 3 2 3 2 1 2 1
Total 7 25 4 36 1 1 1 3 6 1 4 4 4 2 1 2

Abbreviation Full-Form Abbreviation Full-Form Abbreviation Full-Form


CS Critical Speed Lat A Lateral Acceleration TTAvoid Time to Avoid a Collision
Decn Longitudinal Deceleration PET Post-Encroachment Time TTC Time to Collision
DRAC Deceleration Rate to Avoid a Crash RDR Required Deceleration Rate TTI Time to Intersection/Stop-line
DTI Distance to Intersection/Stop-line RS Relative Speed/Speed Difference YR Yaw Rate
DVM Driving Volatility Measures TA Time to Accident

Table 4(b). Conflict measures used at unsignalized intersections (2010-19): Mixed measures and combination of measures
Mixed Measures Combination of Measures
Conflict Type CE Coll. Conf. EA ISS ∆KE NC RECP RI TA, Conf. S TTC, PET,
Prob. Prob. TA
All types 1 1 2
Angle 2 1 1 1 4 1 1
Rear-end 1 1
RTOD 1 1 2
Total 1 2 1 1 2 1 8 1 1 2 1

Abbreviation Full-Form Abbreviation Full-Form Abbreviation Full-Form


CE Collision Energy Conf. S Conflict Speed ∆KE Kinetic Energy Loss per Unit Mass
Coll. Prob. Collision Probability EA Evasive Actions NC Near Crash
Conf. Prob. Conflict Probability ISS Injury Severity Score RECP Rear-end Conflict Potential
RI Risk Indicator
71

Table 5. Conflict measures used at roundabouts (2010-19)


Spatial
Mixed Combination of
Temporal Proximity Measures Proximity Kinematic Measures
Measures Measures
Measures
Conflict Type CT Gap PET TIT TTB TTC PSD BTN CPI Decn Delta- DRAC CE NC

Conf. S
S

TTC,
TA,

LR
All types 2 8 20 1 2 1 1 2 2 1

Rear-end 1 1 1 3 1 1 1 1
Run-off Road 1
Total 1 2 9 1 1 23 1 1 1 1 2 2 1 2 2 1

Abbreviation Full-Form Abbreviation Full-Form Abbreviation Full-Form


BTN Brake Threat Number Decn Longitudinal Deceleration PSD Proportion of Stopping Distance
CE Collision Energy DRAC Deceleration Rate to Avoid a Crash TIT Time-Integrated TTC
CPI Crash Potential Index LR Loom Rate TTB Time to Brake
CT Critical Time NC Near Crash TTC Time to Collision
Conf. S Conflict Speed PET Post-Encroachment Time
72

Table 6(a): Conflict measures for vulnerable road users at intersections (2010-19): Temporal and spatial proximity measures
Temporal Proximity Measures Spatial Proximity Measures
Conflict Type Avbl. RT DST Gap PET TA TTC TTI DSPP DTC Lat D Long D Saf. D
Vehicle-Cyclist 2 10 2 17 2 2 1 1
Vehicle-Pedestrian 5 8 19 1 26 1 2 2
Pedestrian-Cyclist 1 1
Total 2 5 8 29 3 43 2 1 2 3 3 1

Abbreviation Full-Form Abbreviation Full-Form Abbreviation Full-Form


Avbl. RT Available Reaction Time Lat D Lateral Distance TA Time to Accident
DSPP Distance between Stop Position and Long D Longitudinal Distance TTC Time to Collision
Pedestrian
DST Deceleration to Safety Time PET Post-encroachment Time TTI Time to Intersection/Stop-line
DTC Distance to Collision Saf. D Safety Distance

Table 6(b): Conflict measures for vulnerable road users at intersections (2010-19): Kinematic, mixed, and combination of measures
Kinematic Measure Mixed Measures Combination of Measures
Conflict Type ART BRT Cyc. S Decn RS DRAC Max SF Max S Ped Veh. 85th NC PSCI SI EA 85th TTC, Veh. S, TA,
SP S S/PET Max PET, Long D Conf.
SF, EA S
Jerk
Vehicle-Cyclist 1 1 2 1 1 1 1 3 6 1 2
Vehicle-Pedestrian 1 1 1 2 1 1 1 2 1 1 1 9 1 2 3
Pedestrian-Cyclist 1 2
Total 2 2 1 3 3 2 1 1 1 3 1 3 1 1 17 1 1 2 5

Abbreviation Full-Form Abbreviation Full-Form Abbreviation Full-Form


ART Accelerator Release Time Max SF Maximum Absolute Rate of Change in Step Frequency RS Relative Speed/Speed Difference
BRT Brake Response Time Max S Maximum Speed 85th S 85th Percentile Speed
Conf. S Conflict Speed NC Near-crash SI Severity Index
Cyc. S Cyclist Speed PET Post-encroachment Time TA Time to Accident
Decn Longitudinal Deceleration Ped SP Pedestrian Speed Profile Veh. S Vehicle Speed
DRAC Deceleration Rate to Avoid a Crash PSCI Pedestrian Safety Conflict Index Veh S, Long D Vehicle Speed-Distance to Pedestrian
EA Evasive Actions
73

Table 7. Descriptive statistics of the conflict measure thresholds

Signalized Intersections Unsignalized Intersections Roundabouts


Conflict Measures Units
Mean Median Minimum Maximum Mean Median Minimum Maximum Mean Median Minimum Maximum
Brake Reaction Time (BRT) s 0.3 0.3 0.3 0.3 - - - - - - - -
Brake Threat Number (BTN) - 0.5 0.5 0.5 0.5 - - - - 0.5 0.5 0.5 0.5
Collision Probability (Coll. - - - - - 0.05 0.05 0.05 0.05 - - - -
Prob.)
Conflict Probability (Conf. - - - - - 0.03 0.03 0.03 0.03 - - - -
Prob.)
Critical Speed (CS) m/s - - - - 18.77 20.58 13.73 22 - - - -
Critical Time (CT) s - - - - - - - - 4.65 4.65 4.65 4.65
Longitudinal Deceleration m/s2 4.18 4.39 2 5 4.5 4.5 4 5 3.33 3.33 2.22 4.44
(Decn)
Deceleration Rate to Avoid a m/s2 2.12 1.5 1.5 3.35 3.25 3.25 1.5 5 3.35 3.35 3.35 3.35
Crash (DRAC)
Distance to Intersection (DTI) m - 2 2 2 2 - - - -
Encroachment Time (ET) s 0.3 0.3 0.3 0.3 - - - - - - - -
Gap Time (Gap) s 3.05 3.05 2 4.1 4.11 4.24 3 5.1 5.31 5.31 4.12 6.5
Gap: First Lane s 8.3 8.3 8.3 8.3 - - - - - - - -
Gap: Second Lane s 7.1 7.1 7.1 7.1 - - - - - - - -
Integrated Conflict Risk m/s2 1 1 1 1 - - - - - - - -
Index: DRAC component
(ICRI: DRAC)
Integrated Conflict Risk s 4 4 4 4 - - - - - - - -
Index: TTC component
(ICRI: TTC)
Injury Severity Score (ISS) - - - - - 0 0 0 0 - - - -
Kinetic Energy Loss per Unit J - - - - 33.33 0 0 100 - - - -
Mass (∆KE)
Lateral Acceleration (Lat A) m/s2 5 5 5 5 5 5 5 5 - - - -
Modified Time to Collision s 1.25 1 0.88 3 - - - - - - - -
(MTTC)
Post-Encroachment Time s 3.1 3.8 0.8 6 4.11 5 1 5 3.17 2.5 1.67 5
(PET)
74

Signalized Intersections Unsignalized Intersections Roundabouts


Conflict Measures Units
Mean Median Minimum Maximum Mean Median Minimum Maximum Mean Median Minimum Maximum
Proportion of Stopping - - - - - - - - - 1 1 1 1
Distance (PSD)
Relative Acceleration (RA) m/s2 0.97 0.97 0.97 0.97 - - - - -
Required Deceleration Rate m/s2 - - - - 3.58 3.58 3.58 3.58 -
(RDR)
Time Headway (TH) s 2 2 2 2 - - - - -
Time to Brake (TTB) s 1 1 1 1 - - - - 1 1 1 1
Time to Collision (TTC) s 2.14 1.6 0.23 6 1.88 1.5 0.5 5 1.52 1.5 0.5 4
Difference to Time to s - - - - 2.5 2.5 2.5 2.5 - - - -
Collision and Time to Avoid
a Collision (TTC-TTAvoid)
Loom-gated Time to s 2 2 2 2 - - - - - - - -
Collision (TTC, LR)
Yaw Rate (YR) ° 4 4 4 4 4 4 4 4 - - - -
75

Table 8. Parameter estimates of ordinary least squares linear regression models for conflict thresholds

Unsignalized
Signalized Intersections
Intersections
Time-to-collision Post-encroachment Time-to-collision
Variable
(seconds) time (seconds)
(seconds)
Parameter t-stat Parameter t-stat Parameter t-stat
Intercept 3.542 8.86 2.796 5.41 3.00 8.10

Organized traffic indicator


(1 if organized traffic environment, 0 otherwise) -1.047 -2.47 1.568 2.83 -1.566 -3.93

Simulation study indicator - -


(1 if the study was based on microsimulation, 0 otherwise) -1.071 -2.79 - -

Crash-conflict relationship indicator - -


(1 if crash-conflict relationship is the primary purpose of the
study, 0 otherwise - - -1.672 -2.94

R2 0.24 0.574 0.436

Number of Observations 39 29 27
76

Appendix A: List of Reviewed Conflict Measures (see Appendix B for a definition of terms)

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Aggregate An aggregate of the 𝐶𝑃𝑀 = 𝑃(𝑅𝑇 ≥ 𝑇𝑇𝐶) Wang and Suitable for Not widely adopted;
Conflict Conflict Propensity Metric + 𝑃(𝑅𝑇 < 𝑇𝑇𝐶) Stamatiadis application in a
Requires reaction
Propensity values for an intersection × 𝑃(𝑀𝐴𝐷𝑅 < 𝐷𝑅𝐴𝐶|𝑅𝑇 < 𝑇𝑇𝐶) (2013) microsimulation
time and vehicle
Metric (ACPM) by conflict type over all environment;
the road users. 𝐴𝐶𝑃𝑀 = ∑ 𝐶𝑃𝑀 braking capacity
∀𝑖,𝑗 Has demonstrated information for
where, 𝑖, 𝑗 are conflicting pairs of vehicles scalability for large- calculation, which is
RT is assumed lognormal distributed while MADR is scale assessment not readily available;
assumed to follow a truncated normal distribution Difficult to use for
field-based studies
Brake Reaction The time between the 𝑡𝑡 − 𝑡𝑏 Preuk et al. Suitable for Cannot
Time (BRT) instant at which the road (2016) naturalistic methods independently define
user identifies a threat and of observation; a conflict;
the instant it applies the
Can allow driver Not suitable for
brake.
behaviour analysis facility-based
regarding perception- observations;
reaction failure and
Does not measure
aggression
the severity of the
conflict;
Weak correlation
with crashes
Brake Threat Assesses the risk of Gelso and Suitable for Requires the
Number (BTN) collision by the ratio of 𝐷𝑅𝐴𝐶 Sjöberg (2017) application in knowledge of
needed deceleration to naturalistic studies; MADR, which is not
avoid a collision and the 𝑀𝐴𝐷𝑅 generally available;
Suitable for
maximum achievable
application in a Cannot be used in
deceleration for an evasive
field studies based
action (braking)
on facility-based
77

Conflict
Definition Formula Reference Advantages Disadvantages
measure
microsimulation observation of
environment; conflicts
Suitable for
application in ITS-
based collision
avoidance/warning
systems
Change of The ratio of the tangent 𝑑𝜃 Wei et al. Sudden change of Cannot
Curvature angle to the arc length 𝑑𝑠 (2019) curvature indicates independently
(Curv.) traversed by a turning where, 𝜃 is the heading angle and s is the arc length sharply turning describe a conflict
vehicle vehicle that can lead
Relationship with
to conflicts;
crashes is not
Can describe evasive validated
actions taken by
turning vehicles
Collision Energy The total kinetic energy 𝐿 1 1 Astarita and Suitable for It is hard to measure
𝑍𝑡−𝜃−100/3 = . 2. 𝑚(𝑣 sin 𝜃)2 .
(CE) released by a potential 𝑣. 𝑡 2 3 Giofré (2019) application in a the distraction time
crash that is calculated by microsimulation of the drivers in all
where, 𝑡 is the distraction time, 𝜃 is the angle of
explicitly accounting for environment; possible situations;
deviation, 𝐿 is the length of the road segment, 𝑚 is
the distraction time of the
the vehicle mass, and 𝑣 is the vehicle speed. 100/3 Specifically captures Due to the
driver and the angle of
denotes a 1/3 probability that the vehicle will follow a the element of driver probabilistic nature
deviation from a straight
straight or a deviated (±θ) trajectory. error; of the measure, it is
trajectory
computationally
Enables simulation of
expensive to
possible outcomes of
calculate for high
a crash event;
traffic flows
Can model the
outcomes of single-
vehicle crashes
78

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Collision A measure of crash 𝑇𝑇𝐶 Liu et al. Simplifies No information
𝑒− 𝛿
Probability opportunity expressed as a (2013), Chen aggregation of regarding the nature
(Coll. Prob.) monotonically decreasing where, 𝛿 is a scale parameter et al. (2018) conflict severity by of the scaling
function of a proximity normalising the TTC parameter is
measure (e.g., TTC) on (or PET) values; available;
the domain [0,1]
Conveys a distinct Relationship with
idea of the crashes has not been
seriousness of a validated
traffic conflict;
Suitable for network-
wide large-scale
studies
𝑁
Conflict A model used for Zhou et al. Combines measures The form of CHL
Hazardous assessing the safety levels 𝐶𝐻𝐴𝑀 = ∑ 𝐶𝑇𝑖 𝐶𝐻𝐿𝑖 (2011) of both crash models for motorised
Assessment of a traffic facility based 𝑖=1 opportunity and and non-motorised
Model (CHAM) on two aspects: number of severity in one participants is the
where, 𝐶𝑇 is the number of severe conflicts
severe conflicts and metric; same
(determined based on 85th percentile of TTC), 𝑖 is the
Conflict Hazardous Level
traffic conflict type, and 𝐶𝐻𝐿 is the Conflict Explicitly considers TTC is used in CT
(CHL) calculated in the
Hazardous Level of ith conflict type influenced by the conflicts estimation, which
event of a traffic conflict
conflict angles, velocity and the weight of different including non- does not account for
between mixed-traffic
traffic modes motorised modes the mixed traffic
modes
conditions

Conflict The increase in travel time 𝐷𝑖 Shen et al. Attempts to model Provides no
Intensity (CI) in a given lane attributable 𝑇0𝑖 (2013) the effect of traffic information
to traffic conflicts conflicts on traffic regarding conflict
where, 𝐷𝑖 is the total travel delay in lane 𝑖 due to efficiency and crash
traffic conflicts and 𝑇0𝑖 is the free-flow travel time in mechanism
that lane
79

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Conflict The intersection area is 𝑃𝑟𝑜𝑏𝑖𝑗𝑛 × 𝑃𝑟𝑜𝑏𝑖𝑗(𝑛+1) Ma et al. Suitable for Calculation of
Probability divided into numerous (2018) intersection-level constituent
(Conf. Prob.) cells of equal area, and the where, 𝑃𝑟𝑜𝑏𝑖𝑗𝑛 and 𝑃𝑟𝑜𝑏𝑖𝑗(𝑛+1) are the probabilities safety assessment probabilities is based
conflict probability is the of simultaneous arrivals of vehicles 𝑛 and 𝑛 + 1, applications on macroscopic flow
probability that two or respectively, at the cell 𝑖𝑗 variables like
more vehicles will arrive volume and
at the same cell occupancy, hence
simultaneously cannot describe
microscopic conflict
mechanism;
Limited applicability
Conflict In a group of drivers, the Incorporated in ACPM
Propensity crash propensity is the
Metric (CPM) sum of the percentage of
drivers having a reaction
time greater than equal to
TTC and the percentage of
drivers having RT<TTC
but also a maximum
available braking rate
lesser than the required
braking rate.
Conflict Speed The mean speed of the 𝑡
𝑓 Saunier et al. Can be combined Cannot
∑𝑡=𝑡 𝑣
𝑖
(Conf. S) subject road user involved (2011) with other metrics to independently define
in a conflict 𝑇 give a holistic picture a conflict;
of the evolving
Weak correlation
conflict
with crashes
Crash Potential The probability that the Guido et al. Probabilistic Depends on the
Index (CPI) following vehicle’s 𝑡𝑓 (2011) measure, hence can knowledge of the
DRAC exceeds the ∑𝑡=𝑡 𝑃(𝐷𝑅𝐴𝐶 ≥ 𝑀𝐴𝐷𝑅)∆𝑡𝑏 account for vehicle underlying
𝑖
Maximum Available 𝑇 distribution of
80

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Deceleration Rate and driver-level MADR, which is not
(MADR) or the braking heterogeneity; readily available;
capacity of the following
Suitable for Weak correlation
vehicle.
application in with crashes
naturalistic studies;
Suitable for
application in a
microsimulation
environment;
Suitable for
application in ITS-
based collision
avoidance/warning
systems
Critical Speed The speed of a subject 𝑎 Paul and Speed corresponding Requires knowledge
𝑃𝐸𝑇 ×
(CS) vehicle required for 0.039 Ghosh (2018) to critical PET values about the critical
successful braking when a can be calculated deceleration rate of
conflicting vehicle is beforehand and the vehicles, which
already present in the compared with field- is difficult to obtain;
conflict zone measured speeds to
Relationship with
quickly identify
crashes is not
conflicts;
validated
Can support
implementation of
real-time risk
management
measures
Critical Time The time lag from the - Pecchini et al. Specifically accounts Limited
(CT) passing of the nearest (2014) for single-vehicle applicability;
circulating vehicle in a
roundabout to the
81

Conflict
Definition Formula Reference Advantages Disadvantages
measure
approach of the entering run-off-road type Relationship with
vehicle. Higher time lags conflicts crashes is ambiguous
allow for higher speeds
that can result in run-off-
road type crashes.
Deceleration Difference between the So et al. Simple to measure; Cannot
Rate Difference deceleration rates of the (2015) independently define
𝑑2 − 𝑑1 Can be used in all
(DRD) following and leading a conflict;
types of studies;
vehicles
Weak correlation
Suitable for
with crashes
application in ITS-
based collision
avoidance/warning
systems
Deceleration The minimum rate at Guido et al. The most popularly Weak correlation
Rate to Avoid a which a vehicle must 𝑣2 − 𝑣1 (2011) used acceleration- with crashes;
Crash (DRAC) decelerate to avoid a 𝐷𝑅𝐴𝐶𝑟𝑒𝑎𝑟 = based measure;
2𝑑 2 Both conflict and
probable collision and is
2[𝑣2 𝑡1′ −𝑑 ′] Unlike TTC, crash thresholds are
considered an appropriate
𝐷𝑅𝐴𝐶𝑎𝑛𝑔𝑙𝑒 = accounts for sudden unclear;
measure for detecting 𝑡1 ′
disturbances in traffic
dangerous driving Only dependent
𝑑′ 𝑑′ that can lead to
manoeuvres ∀ ≤ 𝑡1 ′ ≤ 2 upon vehicle
𝑣2 𝑣2 conflicts;
capabilities;
𝑑 ′ + 𝑙1 Can be utilised along discounts driver and
𝑡1 ′ = with proximity pavement
𝑣1
measures to capture characteristics
the complete unsafety
picture;
Suitable for
application in
naturalistic studies;
82

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Suitable for
application in a
microsimulation
environment;
Can describe the
evolution of conflicts
Dilemma Zone The probability for an 1, 𝑖𝑓 𝑡ℎ𝑒 𝑣𝑒ℎ𝑖𝑐𝑙𝑒 𝑖𝑠 𝑖𝑛 𝑟𝑖𝑠𝑘 𝑧𝑜𝑛𝑒 𝑎𝑡 𝑦𝑒𝑙𝑙𝑜𝑤 𝑜𝑛𝑠𝑒𝑡 Li et al. (2013) Among the few Does not account for
{
Conflict approaching vehicle in the 𝑃𝑟𝑜𝑏{𝑅𝐸} + 𝑃𝑟𝑜𝑏{𝑅𝐴}, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 measures focussing the pedestrian-related
Potential (DCP) dilemma zone during the exclusively on conflicts resulting
The exact formulation depends on risk scenarios (see
yellow interval to have dilemma zone related due to red-light
reference), one risk scenario is given below:
potential traffic conflicts conflicts; running;
with the leading vehicle or 𝑃𝑟𝑜𝑏{𝑅𝐸}
Comprehensively Several probability
with the opposing traffic = [1 − 𝑃𝑆𝑢𝑏 (𝑆𝑡𝑜𝑝|𝑂𝑍)] accounts for both calculations are
in the next phase × [𝑃𝑙𝑒𝑎𝑑 (𝑆𝑡𝑜𝑝|𝑂𝑍). 𝑃𝑙𝑒𝑎𝑑 (𝐴𝑏𝑟𝑢𝑝𝑡|𝑂𝑍, 𝑆𝑡𝑜𝑝)] rear-end and angle required to arrive at
𝑃𝑟𝑜𝑏{𝑅𝐴} = { 𝑃𝑆𝑢𝑏 (𝑅𝑢𝑛𝑅|𝑂𝑍, 𝐺𝑜). [1 conflicts; the potential, which
− 𝑃𝑆𝑢𝑏 (𝑆𝑡𝑜𝑝|𝑂𝑍)]} may increase the
Takes into account
× [1 − 𝑃𝑙𝑒𝑎𝑑 (𝑆𝑡𝑜𝑝|𝑂𝑍)] complexity of
actions of both the
estimation
where, 𝑃𝑟𝑜𝑏{𝑅𝐸} is the probability of a rear-end leading and the
conflict; 𝑃𝑟𝑜𝑏{𝑅𝐴} is the probability of a right angle subject vehicles
conflict; 𝑃𝑆𝑢𝑏 (𝑆𝑡𝑜𝑝|𝑂𝑍) is the probability of the
subject vehicle in the option zone stopping before the
stop line at the onset of yellow; 𝑃𝑙𝑒𝑎𝑑 (𝑆𝑡𝑜𝑝|𝑂𝑍) is the
probability of the leading vehicle in the option zone
stopping before the stop line at the onset of yellow;
𝑃𝑙𝑒𝑎𝑑 (𝐴𝑏𝑟𝑢𝑝𝑡|𝑂𝑍, 𝑆𝑡𝑜𝑝) is the probability of the
stopping leading vehicle in the option zone making an
abrupt stop at the onset of yellow;
𝑃𝑆𝑢𝑏 (𝑅𝑢𝑛𝑅|𝑂𝑍, 𝐺𝑜) is the probability of a subject
vehicle running the red light given that it is in the
option zone and it chooses to go.
83

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Distance to The distance from the - Noble et al. Simple metric that is Cannot
Intersection / front bumper of the (2016) easy to independently define
Stop Line (DTI) subject vehicle to the measure/calculate; a conflict;
leading edge of the stop
Can be combined Carries little
line
with other conflict information
measures to describe regarding crash risk
the evolution of a
conflict;
Driving Variations in The volatility indices that have been used are standard Wang et al. Suitable for Not suitable for
Volatility-based instantaneous driving deviation, coefficient of variation, mean absolute (2015), Liu application in a microscopic safety
Measures factors (such as speed, deviation, quartile coefficient. of variation, count of and Khattak connected-vehicle assessment;
(DVM) acceleration, vehicular extreme values, and Time-varying Stochastic (2016), environment;
Provide little insight
jerk and/or steering angle). Volatility (see references in the next column for more Kamrani et al.
Allows for into the crash
details) (2017),
macroscopic safety mechanism;
Kamrani et al.
assessment;
(2018), Wali Have not been used
et al. (2018), Suitable for facility- in studies other than
Arvin et al. based safety those focussed on
(2019) assessment; safety assessment
through
Demonstrated
instrumented
relationship with
vehicles
crashes
Encroachment Time during which an Baratian- Simple conflict Relationship with
Time (ET) offending vehicle Ghorghi et al. measure; crashes is unknown;
𝑡𝑓 − 𝑡𝑖
(2016)
infringes upon the Easy measurement; PET is found to be a
right-of-way of better indicator of
Directly observable
crash occurrence
another vehicle
than ET
Evasive Actions Includes actions such as - Hutton et al. Same as Near Crashes (NC)
(EA) turning or opposing (2015)
vehicle quickly changing
84

Conflict
Definition Formula Reference Advantages Disadvantages
measure
path or braking suddenly
to avoid collision during a
turning manoeuvre
Extended Delta- Change in the vehicle’s 𝑣𝑜1 − 𝑎1 𝑡, 𝑖𝑓 (𝑣𝑜1 − 𝑎1 𝑡) ≥ 0 Laureshyn et Shares the Relationship with
V (Ext. Delta-V) velocity because of a 𝑣1 = { al. (2017) advantages of Delta- crash severity is
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
collision if the vehicles V; unvalidated;
𝑣𝑜2 − 𝑎2 𝑡, 𝑖𝑓 (𝑣𝑜2 − 𝑎2 𝑡) ≥ 0
brake at a constant 𝑣2 = { Refinement of the Absence of
deceleration rate from 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
where, 𝑣𝑜1 and 𝑣𝑜2 are the initial speeds (m/s) constant velocity guidelines regarding
conflict detection until the
assumption of Delta- thresholds for
assumed collision. of the subject and conflicting vehicles, V various severity
respectively, at the start of the conflict, levels
t = duration of conflict (s).
Then ∆𝑣 is calculated normally
Gap Time (Gap) In a leading-following Same as TH Huang et al. Shares the Calculation requires
situation, it is the time (2013), Pawar advantages of TH prediction of
between the crossing of and Patil and TTC movement at
the rear of the leader and (2017) constant velocities;
the front of the follower of
Rarely used in
a point on the road
conflict studies as
more pertinent
information is
available from PET
Injury Severity The expected occupant Sobhani et al. Suitable for Occupant injury
Score (ISS) injury severity of a 𝐼𝑆𝑆 = ∑ (𝛼0 + 𝛼𝑖 𝑥𝑖 ) (2013) microscopic severity can be
∀𝑖
conflict, measured using where, 𝑥𝑖 are the independent variables (Kinetic simulation studies; different for various
the estimated kinetic Energy and the type of impact) occupants within the
Considers the point
energy of the subject same vehicle, based
of impact and
vehicle (KE) and the on their seating,
presence of driver
impact type of the which is not
reaction in
expected crash, i.e., near- considered;
estimation, which is
typically not
85

Conflict
Definition Formula Reference Advantages Disadvantages
measure
side, far-side, front, or rear available for delta-v Several assumptions
impact calculations are required for
calculation of the
Kinetic energy of the
crash;
The assumed linear
relationship of ISS
may not be
optimum;
Fails to provide
information
according to typical
injury severity levels
Integrated The Euclidean distance of 2 Li et al. (2017) Attempts to combine Uses DRAC as the
Conflict Risk a point from the origin in √𝑇𝑇𝐶(𝑂𝑖 )2 + 𝐷𝑅𝐴𝐶(𝑂𝑗 ) the indicators of severity indicator,
Index (ICRI) a two-dimensional space crash occurrence and which does not have
defined by indicators of where, 𝑇𝑇𝐶(𝑂𝑖 ) are the TTC value corresponding to severity; a valid relationship
crash occurrence (e.g., discrete risk levels 𝑖 = 0,1,2,3, 𝑎𝑛𝑑 4; and 𝐷𝑅𝐴𝐶(𝑂𝑗 ) with crash severity;
are the DRAC values corresponding to discrete risk Defines crash risk
TTC) and severity (e.g.,
levels j= 0,1,2,3, 𝑎𝑛𝑑 4. The risk levels are defined in according to several Discrete risk levels
DRAC)
the referenced study. risk levels to model fail to capture the
the entire safety continuous evolution
continuum of risk with time
Jerk The rate of change of (Tageldin et Can describe the The evidence of the
acceleration with respect 𝑎𝑓 − 𝑎𝑖 al., 2015, Wei severity of the relationship with
to time et al., 2019, conflict event; crashes has not been
𝑡𝑓 − 𝑡𝑖
Huertas-Leyva demonstrated for
Can be used in all
et al., 2019) intersections;
types of studies;
Due to higher
Suitable for
amount of
application in
calculations involved
naturalistic studies;
if only speed data is
86

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Can be utilised along available, the
with proximity calculation of jerks
measures to capture is considerably
the complete unsafety affected by the noise
picture; in the measured raw
data
Has a positive
relationship with
crashes
Kinetic Energy The kinetic energy 1 1 Ma et al. Provides a simple Limited
Loss per Unit ∆𝐾𝐸 = (𝑣1 2 + 𝑣2 2 ) − 𝑣1 𝑣2 cos 𝜃
released because of a 4 2 (2018) estimate of crash applicability;
Mass (∆KE) collision between vehicles severity;
Ignores loss of
of comparable masses.
Can be combined energy due to heat,
with conflict sound, and physical
probability measure deformation of the
to provide the overall vehicles;
safety index of an
Predicts future
intersection
trajectory based on
constant velocity
from the time of
measurement
Lateral The instantaneous Guo et al. Suitable for Cannot
Acceleration / acceleration/deceleration 𝑣𝑓,𝑙𝑎𝑡 − 𝑣𝑖,𝑙𝑎𝑡 (2010) application in independently define
Deceleration of the vehicle in the lateral naturalistic studies; a conflict;
𝑡𝑓 − 𝑡𝑖
(Lat A / Lat D) direction of motion
Accounts for the Weak correlation
steering actions taken with crashes
as part of collision
evasion
Left-turn Driver The product of the 𝑝𝑔𝑎𝑝 {𝑟𝑔𝑎𝑝 } × 𝑝𝑎𝑐𝑐 {𝑟𝑔𝑎𝑝 } × 𝑝𝑑𝑎𝑟 {𝑟𝑔𝑎𝑝 } Santiago- Captures the Limited application;
Safety Index probabilities of a Chaparro et al. complete causal
(LDSI) particular gap occurring in (2010) chain of events for
87

Conflict
Definition Formula Reference Advantages Disadvantages
measure
the conflicting traffic where, 𝑟𝑔𝑎𝑝 is a given gap in the conflicting stream, turning movements- Depends on manual
stream, the subject vehicle 𝑝𝑔𝑎𝑝 is the probability of occurrence of 𝑟𝑔𝑎𝑝 , 𝑝𝑎𝑐𝑐 is related conflicts; observation of
accepting that gap, and the probability that 𝑟𝑔𝑎𝑝 is accepted by the left-turning adverse reactions;
Combines the
occurrence of an adverse driver, and 𝑝𝑑𝑎𝑟 is the probability of observing an elements of traffic Relationship with
driver reaction (e.g., adverse reaction by the conflicting vehicle efficiency and safety crashes is not
diving of the vehicle nose,
into one metric validated
sudden acceleration or
deceleration by the
opposing vehicle) by the
driver of the conflicting
vehicle.
Longitudinal The instantaneous Songchitruksa Easy measurement; Reactive measure;
Acceleration / acceleration/deceleration 𝑣𝑓,𝑙𝑜𝑛𝑔 − 𝑣𝑖,𝑙𝑜𝑛𝑔 and Zha
Suitable for Cannot
Deceleration of the vehicle in the (2014)
𝑡𝑓 − 𝑡𝑖 application in independently define
(Accn / Decn) longitudinal direction of
naturalistic studies; a conflict in
motion.
proactive
Good correlation of
A hard assessment;
hard
braking/acceleration event
braking/acceleration Arbitrary setting of
is identified when this
events with crashes at the thresholds to
value is above a certain
intersections define hard
threshold.
braking/acceleration
event
Loom Rate The loom rate of a point is (𝐩𝟐 − 𝐩𝟏 ) × 𝐯̂𝟏 + (𝐩𝟐 − 𝐩𝟏 ) × 𝐯𝟐 Ward et al. Describes the vehicle Cannot
its angular velocity from ‖𝐩𝟏 − 𝐩𝟐 ‖ (2015) interactions in a 2-d independently define
the point of view of the plane, hence can a conflict;
ego point, pi. This ego 𝐯̂𝟏 = 𝐯𝟏 + (𝐩𝟏 − 𝐩𝐜 ) × 𝛚𝟏 capture all types of
Requires accurate
point is known as the conflicts when
knowledge of
loom point. combined with a
geometries of the
suitable conflict
conflicting vehicles,
measure;
which may be hard
to ascertain from
88

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Suitable for commonly used
application in a detectors in field
microsimulation studies
environment;
Can help describe the
evolution of conflicts
Modified Time A modification on TTC −∆𝑠 ± √∆s 2 + 2𝑑∆𝑎 Essa and Overcomes the Involves calculation
to Collision that estimates TTC based ∆𝑎 Sayed (2019) constant velocity of more parameters
(MTTC) on relative speed and limitation of TTC; than TTC, hence
relative acceleration of the more chances of
Estimates the
interacting vehicles incorrect estimation;
collision risk even
when the following Hard to physically
vehicle speed is interpret in terms of
lower than the relationship with
following vehicle driver reaction times
speed
Near Crash (NC) A circumstance requiring Varies with study and application; see references Guo et al. Suitable for No consistent
rapid evasive manoeuvres (2010), Wu application in definition of what
by the subject vehicle, or and Jovanis naturalistic studies; triggers to be used
the conflicting vehicle, to (2012), and their thresholds;
Suitable for network-
avoid a crash in such a Muralidharan
wide large-scale Requires long
way that the undertaken et al. (2016)
studies; durations for
manoeuvre causes the and several
collection of
vehicle to approach the others Comprehensive
adequately sized
limits of its capabilities measurement of
sample;
(e.g., vehicle braking conflicts using
greater than 0.5 g) several metrics; Data collection is
costly owing to the
Validated
use of several
relationship with
instruments;
crashes
89

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Unsuitable for
application in
facility-based
evaluations,
especially using
video observations
Post- The time difference Peesapati et al. Directly observable; Has a single value,
Encroachment between the moment a (2013) so cannot describe
𝑡1 − 𝑡2 No movement
Time (PET) conflicting vehicle passes evolution of the
assumptions are
out of the potential conflict;
required;
collision area and the
More suitable for
moment of arrival at the Very frequently used
angle-type conflicts
potential collision point by in studies;
rather than collinear
the subject vehicle
Has demonstrated (head-on and rea-
possessing the right-of-
scalability for large- end) type conflicts
way.
scale assessment;
More appropriate
than TTC for
angle/crossing
conflicts
Predicted PET / The expected PET value at Same as PET Mohamed and Overcomes the Calculation requires
Time Advantage each moment of the Saunier (2015) limitation of PET prediction of
(pPET/TAdv) interaction if the road regarding continuous movement at
users continue with the examination of constant velocities;
same speeds and paths conflict evolution;
Relationship with
More appropriate crashes is not
than TTC for investigated
angle/crossing
conflicts
90

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Preemptive The ratio of the length of 𝑙𝑃𝑎𝑠𝑠 Xiao et al. A more refined Accurate vehicle size
Level (PL) vehicle 1 (subject vehicle) 𝑃𝐿1 = (2011) version of predicted information is
𝑙1
that has passed the PET giving enhanced required for
potential conflict point 𝑙𝑃𝑎𝑠𝑠 information on crash calculation;
𝑃𝐿2 = −
(PCP), based on predicted 𝑙1 mechanism;
Knowledge of
future trajectory, to the where, 𝑙𝑃𝑎𝑠𝑠 is the length of vehicle 1 that has passed Explicitly takes into drivers’ critical
total length of vehicle 1 the PCP account driver preemptive level is
when vehicle 2
decision-making required to
(conflicting vehicle) is
behaviour determine conflicts;
predicted to reach the
PCP. The PL of vehicle 2 Limited
is the inverse of the PL applicability;
value of vehicle 1.
Relationship with
crashes is not
investigated
Probability of The probability of an Depends on the motion prediction algorithm; see Mohamed and It can distinguish Involves calculating
Unsuccessful evasive action being reference for more details Saunier (2013) between interactions all possible future
Evasive Action unsuccessful calculated as with the same TTC courses of action,
(P(UEA)) the number of predicted value, but where the which can be time
collisions divided by the conflicting vehicles and resource-
product of the number of can perform different consuming;
trajectories predicted for types of potential
Validity depends
the two road users. evasive actions.
upon the accuracy of
Thus, it controls for
the motion prediction
cases of differing
algorithm
possible crash
intensity that have
the same TTC value
Proportion of The ratio of the remaining 𝑅𝐷 Guido et al. Simple to measure; Remaining distance
;
Stopping distance (RD) of the 𝑀𝑆𝐷 (2011) measurement ignores
Provides a natural
Distance (PSD) subject vehicle (to the driver
conflict threshold of
potential collision point) characteristics;
91

Conflict
Definition Formula Reference Advantages Disadvantages
measure
to the minimum safe 𝑣2 1.0 to identify safety- Poor relationship
stopping distance (MSD) 𝑀𝑆𝐷 = critical events with crashes
2𝑑
Rear-end Crash A temporal measure 𝐶𝑎𝑠𝑒 1: 𝑇𝐻 − 𝑙1 − 𝑅𝑇2 − 𝐵𝑇2 < 0 Dimitriou et As the name Limited
Potential calculated based on the {𝐶𝑎𝑠𝑒 2: 𝑇𝐻 − 𝑙1 + 𝐵𝑇1 − 𝑅𝑇2 − 𝐵𝑇2 < 0 al. (2018) suggests, suitable for applicability;
(RECP) headway, the driver's 𝐶𝑎𝑠𝑒 3: 𝑆1 − 𝑆2 − 𝑆 < 0 rear-end conflicts,
Unsuitable for
perception-reaction time particularly in
The subscripts 1 and 2 denote the properties of facility-based
and the brake time. dilemma zone;
vehicle 1 (leading) and vehicle 2 (following), observation studies;
respectively Suitable for network-
Requires driver
wide large-scale
reaction times of
studies;
both drivers, hence
Suitable for even unsuitable for
application in a naturalistic studies
microsimulation
environment;
Suitable for
application in ITS-
based safety
enhancement systems
Relative The maximum relative ∆𝑠 Lee et al. Simple metric that Provides limited
Speed/Speed speed of the conflicting (2013) can be used in information about
Difference (RS) vehicles combination with crash and conflict
other measures to mechanisms, hence
describe conflicts only used in
holistically conjunction with
other metrics
Required The deceleration rate 𝑣 Babu and The measure is Like DRAC, the
Deceleration required to safely stop a 2𝑃𝐸𝑇 Vedagiri analogous to the threshold of conflict
Rate (RDR) vehicle where the time (2017) more popular DRAC determination is
headway between the metric and shares its difficult in this case;
advantages;
92

Conflict
Definition Formula Reference Advantages Disadvantages
measure
subject and the conflicting Due to PET Cannot describe the
vehicle is equal to PET observation evolution of a traffic
requirement, this interaction into a
metric is measured conflict;
after-the-fact (after
Relationship with
the vehicles have
crashes is not
crossed) and
validated
overcomes the
constant velocity
assumption of DRAC
Required Acceleration required by 2(𝑑 − 𝑣𝑡) Sharma et al. As the name Limited
Longitudinal the vehicle to cross the 𝑡2 (2011) suggests, suitable for applicability;
Acceleration stop line prior to the onset conflicts due to red-
where, 𝑑 is the distance from the stop line, and 𝑡 is Does not account for
(RA) of red light running;
the time remaining till the onset of red driver and pavement
Can help model the characteristics
dilemma zone hazard
2 +𝛼 (𝐷𝑇𝐼 +𝑝 )2 ]
Risk Indicator A quantitative measure of 𝑒 −[𝛼1(𝐷𝑇𝐼1+𝑝1) 2 2 2 Kamal et al. Similar to Collision Estimation of
(RI) collision risk computed as (2015) Probability (Coll. collision point is not
a monotonically where, 𝐷𝑇𝐼𝑖 is the distance to the stop line of the Prob.) and Severity straightforward in
decreasing function of the vehicle 𝑖, 𝑝𝑖 is the distance of the potential collision Index (SI) indicators; real-world situations;
respective distances in point from the stop line for vehicle 𝑖, and 𝛼𝑖 are shares their
estimable parameters Ignores
cartesian coordinates of advantages;
heterogeneities due
the two vehicles from the
Suitable for to driver, vehicle,
collision point
microscopic and pavement
simulation studies characteristics
Safety Index An index combining 𝛼∆𝐾𝐸 Wang et al. Combines measures The calculation of
(Saf. I) conflict probability and 𝑒 𝛽𝑃𝐸𝑇 (2019b), of both crash kinetic energy
severity, which is Alhajyaseen opportunity and released in a
computed based on the (2014) severity in one collision requires
released kinetic energy of metric; several assumptions
the hypothetical crash and is cumbersome;
93

Conflict
Definition Formula Reference Advantages Disadvantages
measure
weighted by the Post Can be calibrated to Does not provide
Encroachment Time account for several crash severity
(PET) features affecting information
crash outcome like according to severity
vehicle size, vehicle levels
body design and the
presence of vehicle
safety features such
as airbags;
Demonstrated
relationship with the
number of severe
crashes
Severe Crash Average of the Incorporated in ASCM
Metric (SCM) probabilities of occurrence
of crashes of a given
severity-type. The crash
occurrence is estimated
like in the CPM metric,
while the severity
probability is calculated as
the power model of delta-
V.
Severity Index A unitless measure of ∆2 Autey et al. Simplifies Information about

𝑒 2𝑅𝑇 2
(SI) conflict severity (2012) aggregation of the response time of
(nearness-to-collision) that where, ∆ is any temporal proximity measure like TTC conflict severity by the drivers is
ranges from 0 to 1; 0 or PET normalising the TTC required, which is
corresponds to (or PET) values; not readily available
uninterrupted passages in field studies;
Conveys a distinct
idea of the
94

Conflict
Definition Formula Reference Advantages Disadvantages
measure
seriousness of a Relationship with
traffic conflict; crashes has not been
validated
Suitable for network-
wide large-scale
studies
Stopping Distance covered by a 𝑣2 Oh et al. Can be used as a Conveys no conflict
Distance (SD) vehicle under maximum 254𝜇 (2010) decision tool to information by itself
deceleration predict conflicts
Time Exposed Summation of all 𝑇 (Rahman et Continuous Shares the
TTC (TET) durations (over the ∑ 𝛿(𝑡). ∆𝑡 al., 2019) evaluation of the disadvantages of
considered time period) 𝑡=0 conflict over its TTC;
for which the TTC-value 1 ∀ 0 ≤ 𝑇𝑇𝐶(𝑡) ≤ 𝑇𝑇𝐶 ∗ entire duration is
𝛿={ Cannot assess the
between an interacting possible;
0 𝑒𝑙𝑠𝑒 severity of the
pair of vehicles remains
Suitable for network- conflict;
below the TTC threshold
wide large-scale
for conflicts Cannot
studies;
independently define
Suitable for a conflict
application in a
microsimulation
environment
Time Headway The time between the 𝑡1 − 𝑡2 (Peng, 2018) Very simple to Weak correlation
(TH) front/rear of the leading measure/calculate; with crashes;
vehicle and the front/rear
Actual gap between Cannot
of the following vehicle
vehicles so easy to independently define
crossing a particular point
visualise the safety a conflict;
on the road
risk
Cannot distinguish
between aggressive
driving and conflict
situation
95

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Time Integrated Integral of the TTC-profile 𝑇 (Rahman et Shares the Shares the
TTC (TIT) during the time the TTC- ∫ [𝑇𝑇𝐶 ∗ − 𝑇𝑇𝐶(𝑡)]𝑑𝑡 al., 2019) advantages of both disadvantages of
0
value between an TTC and TET; TTC;
interacting pair of vehicles ∀ 0 ≤ 𝑇𝑇𝐶(𝑡) ≤ 𝑇𝑇𝐶 ∗
Can additionally Cannot
remains below the
indicate the severity independently define
threshold value of TTC.
of the conflict a conflict
Time to The time that remains to a Same as TTC Archer and Wang et al. (2019a) Not very frequently
Accident (TA) collision from the moment Young (2010) state that TA=0 used;
that one of the road users indicates a
Cannot account for
starts an evasive action, if perception-reaction
cases without
they had continued with failure and as such
evasive action;
unchanged speed and can be used in a
directions bivariate setting Calculation requires
along with TTC or prediction of
PET; movement at
constant velocities;
Simple to measure
Cannot
independently define
a conflict;
Cannot distinguish
between aggressive
driving and conflict
situation
Time to Avoid a The duration between the 𝑣2 (Fu et al., Suitable for Requires driver and
+ 𝑅𝑇 + 𝐵𝑇
Collision point at which a potential μg 2018) application in vehicle parameters,
(TTAvoid) collision is detected and naturalistic studies; hence not suitable
the point at which the for field studies with
Suitable for
collision is just avoided facility-based
application in a
observation of
microsimulation
conflicts;
environment;
96

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Suitable for Not suitable for
application in ITS- cases without a
based collision collision course
avoidance/warning
Not widely used
systems

Time to Brake The time period before 𝐷 = (𝑑1 − 𝑑𝑚𝑎𝑥 ){(𝑑2 − 𝑑𝑚𝑎𝑥 )(𝑣1 − 𝑣2 )2 Gelso and Can be used for Unsuitable for large-
(TTB) full braking is needed to + 2𝑆[𝑑2 2 + 𝑑𝑚𝑎𝑥 (𝑑1 − 𝑑2 ) Sjöberg (2017) continuous risk scale application for
avoid a collision assessment; estimating the risk of
− 𝑑1 𝑑2 ]}
entire traffic stream;
Suitable for
(𝑣1 − 𝑣2 )(𝑑1 − 𝑑2 ) + √𝐷 application in Requires numerical
𝑇𝑇𝐵 =
𝑑2 2 + 𝑑𝑚𝑎𝑥 (𝑑1 − 𝑑2 ) − 𝑑1 𝑑2 naturalistic studies; solution, hence,
computationally
where, 𝑑𝑚𝑎𝑥 is the maximum emergency deceleration Suitable for
expensive;
application in ITS-
based collision Absence of
avoidance/warning guidelines regarding
systems the criticality
threshold
Time to The time that remains 𝑑 Sayed et al. Most frequently used Calculation assumes
Collision (TTC) until a collision between ; 𝑖𝑓𝑓 𝑣2 > 𝑣1 (2012) conflict measure with prediction of
∆𝑣
two vehicles if the validated relationship movement at
collision course and speed with crashes; constant velocities,
difference are maintained which is
Has demonstrated
unreasonable;
scalability for large-
scale assessment; Not suitable for
cases without a
Can be used in both
collision course;
naturalistic as well as
facility-based Is undefined in the
methods of conflict case where the
observation: leading vehicle is
moving at a higher
97

Conflict
Definition Formula Reference Advantages Disadvantages
measure
Continuous speed than the
examination of follower;
conflict evolution is
Cannot account for
possible;
the cases with
Suitable for sudden disturbance
application in ITS- in traffic flow
based collision
avoidance/warning
systems
Time to The time that remains 𝑥1 − 𝑥𝑆 Hurwitz et al. Simple estimation of Cannot reliably
Intersection/Stop until a vehicle reaches the 𝑣1 (2012) the possibility of define crash risk;
Line (TTI/TTS) stop line or a point inside conflict;
where, 𝑥1 is the position of the vehicle, 𝑥𝑆 is the Ignores the actions
the intersection if it
position of the stop line, and 𝑣1 is the vehicle speed Suitable for of any conflicting
maintains a constant
prediction of vehicle
velocity
violations such as
red-light running
Yaw Rate (YR) The angular velocity of 𝑑𝜑 Tageldin et al. Accounts for Provides limited
the road-user rotation 𝑑𝑡 (2015) swerving as an information and
around the z-axis or the where, 𝜑 is the heading angle evasive action taken cannot
rate of change of the by some road users, independently define
heading angle. especially a conflict;
motorcycles and
Relationship with
bicycles;
crashes is not
Suitable for validated
application in the
heterogeneous traffic
environment
98

Appendix B: Definitions of terms and notations used in this study

Variable Unit Description


∆a m/s2 Relative acceleration of the pair of interacting vehicles
∆t s Any small, uniform time step between ti and tf
∆s m/s Relative speed of the pair of interacting vehicles
a1 m/s2 Acceleration rate of vehicle 1 (typically subject vehicle, if not mentioned
otherwise)
a2 m/s2 Acceleration rate of vehicle 2 (typically conflicting vehicle, if not
mentioned otherwise)
af m/s2 Acceleration rate of the primary vehicle at the final time of observation
ai m/s2 Acceleration rate of the primary vehicle at the initial time of observation
b A binary variable indicating whether a pair of vehicles
BT s Braking time of the vehicle
d m Longitudinal spatial gap between the rear of the leading vehicle (vehicle
1) and the front of the following vehicle (vehicle 2)
d' m Distance of the closest boundary of conflict area from vehicle 2
2
d1 m/s Deceleration rate of vehicle 1 (typically subject vehicle, if not mentioned
otherwise)
d2 m/s2 Deceleration rate of vehicle 2 (typically conflicting vehicle, if not
mentioned otherwise)
l1 m Length of vehicle 1
2
MADR m/s Maximum available deceleration rate
p1 m Position vector of the ego point (point closest to the other vehicle) of
vehicle 1 in a 2-d plane
p2 m Position vector of the ego point (point closest to the other vehicle) of
vehicle 2 in a 2-d plane
pc m Position vector of the centre of rotation of vehicle 1 in a 2-d plane
RT s Driver’s perception and reaction time
S1 m The distance travelled by vehicle 1 in the time ∆t
S2 m The distance travelled by vehicle 2 in the time ∆t
S m Distance headway between the pair of interacting vehicles
t s Any instant of time
T s Total travel time
t1 s Time taken by vehicle 1 to reach the conflict area
t2 s Time taken by vehicle 2 to reach the conflict area
t1’ s Time taken by vehicle 1 to clear the conflict area
tb s Time at which brake is applied by the primary vehicle
tf s Final time of observation
ti s Initial time of observation
tt s Time of threat identification by the primary vehicle
99

Variable Unit Description


TTC* s Threshold value of TTC for conflicts
v1 m/s Speed of vehicle 1
v1 m/s Velocity vector of the ego point (point closest to the other vehicle) of
vehicle 1 in a 2-d plane
v2 m/s Speed of vehicle 2
v2 m/s Velocity vector of the ego point (point closest to the other vehicle) of
vehicle 2 in a 2-d plane
vi,lat m/s Initial speed of the vehicle in the lateral direction
vf,lat m/s Final speed of the vehicle in the lateral direction
vi,long m/s Initial speed of the vehicle in the longitudinal direction
vf,long m/s Final speed of the vehicle in the longitudinal direction
μ Coefficient of friction of the pavement surface
g m/s2 Acceleration due to gravity
ω1 radians/s Yaw rate of the ego point of vehicle 1 in a 2-d plane

View publication stats

You might also like