Professional Documents
Culture Documents
Abstract: Pedestrian crash data are not available in sufficiently large quantity and suffer from known problems such as a low-mean small
sample, underreporting, and misclassification. Moreover, under heterogeneous traffic conditions, due to the involvement of multiple classes
of vehicles, pedestrian–vehicle interactions become even more complex. To address this limitation, a pedestrian risk index (PRI) linking the
probability of a crash between vehicle and pedestrian and the severity of the conflict was developed for varying road and traffic conditions.
Data were analyzed from nine locations in different parts of India, accounting for variations in geographical distribution and pedestrian–
vehicle interactions. The derived values of PRI were assessed by vehicle type and road geometry. The PRI value was significantly higher
when the approaching vehicle was three-wheelers (3W), two-wheelers (2W), and cars compared with heavy vehicles such as buses and
trucks. This indicates that the severity of conflicts is higher for lighter vehicles. Furthermore, the addition of lanes increases the PRI value.
As an important outcome, variation in PRI values was modeled as a function of vehicle speed, pedestrian volume, and vehicle volume using
a multilinear regression approach. The developed model can enable planners and engineers to compute PRI using the independent variables
and to evaluate pedestrian safety at urban midblock crossings. Overall, this research contributes immensely to assessing the prevailing level
of safety at crosswalks under heterogeneous traffic conditions, thereby improving pedestrian safety. DOI: 10.1061/JTEPBS.0000421.
© 2020 American Society of Civil Engineers.
Author keywords: Safety; Pedestrian; Crosswalk; Midblock; Traffic conflict; Risk index; Surrogate measures.
may hold great potential to reduce the crash frequency and severity
for pedestrians (Jiang et al. 2015). Field traffic data were collected about the type of approaching vehicle, and reduction in the severity
through video recording and image processing at unsignalized of conflicts.
midblock crosswalks in Beijing, China and Munich, Germany. TTC commonly has been implemented as a measure of conflict
Centered vehicle–pedestrian conflict situation, pedestrian speed severity for the whole interaction process. It originally was defined
performance during different walking phases, pedestrian waiting as the time remaining until two vehicles will collide if both
behavior related to waiting decision choice and waiting time, continue at their present speeds along their respective trajectories.
and pedestrian gap acceptance were statistically analyzed to under- Hayward (1972) explained the TTC using a time-space diagram
stand the pedestrian behavior in the conflict process. Intercultural and inferred that if the two vehicle paths remained unchanged, there
comparisons were made between China and Germany. is a maximum probability of a crash. Time to conflict under this
Chung (2018) analyzed the injury severity in taxi–pedestrian condition (collision) becomes zero. However, if one of the vehicles
crashes using measures, such as the TTC, speed, angle, and region changes its path, TTC may increase or decrease. Furthermore, if
of the crash. Five variables were found to have a greater effect on either a pedestrian or vehicle performs evasive action such as stop-
injury severity: crash speed, crashes in no-median section, crashes ping, the TTC tends to infinity. This forms a major limitation of
in which the secondary impact object of pedestrians was the crash using TTC as a risk index. TTC can be detected easily in rear-
vehicle, crashes in which the third impact object of pedestrians was end conflict situations because the trajectories of the paired vehicles
another moving vehicle, and crashes in which the third impact re- are assumed to overlap. However, it cannot be detected (or does not
gion of pedestrians was their head. However, injuries were less se- exist) in most interactions if the trajectories of the paired users do
vere in crashes in which the first impact region on the pedestrian not intersect, for example, in pedestrian–vehicle conflict and con-
was their leg, crashes in which the car moved in a straight line, and flict between left-turn and opposing through vehicles. In rear-end
crashes involving junior high school students. Chaudhari et al. conflict, the following vehicle will collide with the leading vehicle
(2020) evaluated pedestrian–vehicle conflicts at midblock cross- if the speed of the follower is higher. However, in pedestrian–
walks under mixed traffic conditions using different surrogate vehicle conflict, cases in which the pedestrian and the vehicle oc-
safety measures such as PET, conflict ratio, and yielding compli- cupy the trajectory intersection point at the same moment are rare.
ance rate. They correlated PET data and speed of approaching Furthermore, for PET, only their passing times at the conflict point
vehicles to define the severity of the conflict. are necessary. Another important property of PET is that it is con-
Overall, it can be concluded that a substantial literature on the tinuous from crash-free operations to crash occurrences, with a dis-
application of TCTs to assess safety levels exists. Very few studies tinct boundary at zero. A smaller value of PET implies a greater risk
reported on pedestrian–vehicle conflicts and developed an appro- of vehicle–pedestrian crashes. In the context of vehicle–pedestrian
priate measure for pedestrian safety. However, most of the past conflict assessment, PET similarly can be defined as the time
studies are not related to heterogeneous traffic conditions, in difference between the departure of the encroaching pedestrian
which the behavior of road users certainly is different from that from the potential crash point and the arrival of the conflicting ve-
observed in developed countries. Moreover, some critical factors hicle at the crash point, or vice versa. However, because PET con-
such as the type of approaching vehicle, the level of pedestrian siders only the last moment of the interaction, it has limitations in
activity, and non-lane-based traffic movement prevailing in India indicating pedestrian safety, because no information is provided
were not considered in past studies. To fill this research gap, this during vehicle–pedestrian interaction. However, events in which
paper proposes and illustrates the working of a methodology to the approaching vehicle decelerates to a near stop to avoid a crash
assess safety at midblock locations under heterogeneous traffic with the conflicting pedestrian may have PET values that do not
conditions using a proximal surrogate safety indicator. TCT was reflect the true severity of the interaction. These limitations are well
derived based on the driver’s actual behavior, particularly under accounted for in the calculation of PRI because it effectively covers
heterogeneous traffic conditions at unmarked and marked cross- different phases of a pedestrian–vehicle interaction. Moreover,
walks in India. TTC of both the pedestrian and the vehicle is considered in defining
The remainder of the paper is organized in seven sections. A conflict. Therefore, PRI inherently considers TTC in its computa-
comprehensive explanation of PRI and motivation for adopting tion, and thus it can be considered comprehensive.
PRI are explained in the section “Pedestrian Risk Index.” A case
study, data collection, and data extraction are illustrated in the sec-
Deriving Pedestrian Risk Index
tion “Case Study.” Section “Analysis and Results” comprehen-
sively explains the variation in conflict probability and PRI by A pedestrian approaching a crossing area may stop and yield if the
vehicle type and road geometry. As an important outcome, the approaching vehicle is moving and is too close to the crossing area.
PRI is modeled as a function of pedestrian volume, vehicular speed, In other situations, the driver is too close to the crossing area and
and vehicular volume using a multilinear regression technique, does not have adequate time to slow down and come to a complete
which is explained in the section “Application of Study,” followed stop. Therefore, several phases may occur if the vehicle is farther
by the “Conclusion” section. from the crossing area (Fig. 1).
Fig. 1. Conflict representation due to pedestrian and vehicle movement. (Vehicle and bike images by authors.)
Fig. 2. Photographs of different phases: (a) stopping phase; (b) conflict phase; and (c) passing phase. (Images by Avinash R. Chaudhari.)
T r = perception and reaction time of driver (s); DyðvÞ = longitudinal The volume:capacity ratios for the subject study locations varied
vehicle distance from conflict area (m); and V v T r = distance trav- from 0.15 to 0.8, indicating traffic states ranging from free-flow to
elled during perception and reaction time (m). capacity conditions. The pedestrian cross flow varied from 10 to
Furthermore, the difference between the pedestrian time to reach 180 pedestrians=15 min in one crossing direction.
conflict area (TTCp ) and the TTCv at instant i during the passing
phase (TTCp > TTCv ) is ΔT 1 , whereas the difference between
T s and TTCv at instant i during the stopping passing phase Pedestrian–Vehicle Trajectory Data Development
(TTCv > T s ) is ΔT 2 . The trajectory data of a pedestrian and a vehicle in interaction were
Considering the complexity of computing PRI, the following extracted manually from the recorded video (for each interaction
equation was used to compute the PRI (Davis et al. 2003; irrespective of the time of the day or traffic condition). The pres-
Cafiso et al. 2011): ence of multiclass vehicles with poor lane discipline and crossing
pedestrians made it difficult to use image processing based on any
PRI ¼ ΣTTTD ðV 2ImpactSpeed;i ΔT i Þ ð5Þ
automation tools for extracting both vehicle and pedestrian trajec-
where V ImpactSpeed;i = potential collision speed at instant i (m=s); tories. Trajectories of each approaching vehicle and pedestrian
and ΔT i = difference between T si and TTCvi (s). were captured on a two-dimensional coordinate system considering
PRI can be computed individually for each phase of pedestrian– a grid of 1.6 × 5.0 m over a 30-m trap length in the longitudinal
vehicle interaction, i.e., the passing phase, the stopping phase, and direction of the road. This grid size was used because of the
the conflict phase. This study computed PRI for the conflict phase width of the roads and the condition of the crosswalks. The grids
only. were created manually using AutoCAD 2014 software. The grid
then was overlaid on the video using Corel Video Studio ProX9.
Figs. 4(a–d) show crosswalks with overlaid grids at different study
Case Study locations. The pedestrian and the type of vehicle were detected
at the entry point and then tracked frame-to-frame in unmanned
aerial vehicle (UAV) video. In this study, the trajectories of vehicles
Study Area and Data and pedestrians were extracted at intervals of 0.04 s using a detec-
A study was carried out by collecting data at nine midblock loca- tion and tracking system developed in past studies. The system
tions in five different cities of India (Mumbai, Surat, Ahmadabad, mainly includes three steps (Fig. 5): (1) video stabilization; (2) ve-
Chandigarh, and Delhi). The study locations were selected in such hicle detection; and (3) vehicle tracking in case of the vehicle and
a way that capturing of precise movement of the pedestrians and the procedure for a pedestrian. The video is run frame by frame for
vehicles simultaneously was possible. Videographic survey was every object (pedestrian/vehicle). During the UAV motions, after
conducted in the June, October, November, and December of 2015 detection of vehicle type, the object position (coordinates) are
and February of 2016 to capture pedestrian and vehicular move- marked at the entry point then the vehicle is tracked. The trajectory
ment at selected locations during normal working days in ideal of the vehicle is derived after the tracking is finished. Commonly,
weather conditions. Data were recorded using a wide-angle high the entry and exit regions are set at the upstream and downstream of
definition digital video camera (frame rate of 29 frames/s, and a midblock section. When a vehicle enters the entry region (i.e., the
55× zoom), which was mounted at a vantage point on a multistory front bumper of the vehicle crosses the entry line), it is detected and
building in such a way that it recorded the movements of pedestrian tracked. The tracking is finished when the vehicle enters the exit
and vehicles. The camera was not visible to the drivers to avoid region (i.e., the front bumper crosses the existing line). The pedes-
influence on their behavior, and allowed a sufficiently wide zone. trian detection and tracking methodology was developed similarly
The details of the study locations are shown in Fig. 3. Geometric to the extraction of vehicle trajectory. The developed system con-
details, observed traffic and pedestrian flow characteristics, and sists of two components, i.e., a detector and a tracker. When pedes-
average vehicular traffic speed recorded at selected study locations trians are detected, their coordinates (based on the grids) are fed as
are presented in Table 2. The average vehicular traffic flow during inputs for initialization of trackers. After tracking, the pedestrian
the survey at selected locations was between 2,080 and 4,174 trajectory data can be saved for further analysis. The available ob-
passenger car units ðPCUÞ=h, pedestrian flow rate was between servations are trajectory profiles based on time series. From these
484 and 1,235 pedestrians=h, and mean vehicular speed was data, all relevant quantities of vehicles and pedestrians, such as po-
between 23.63 and 38.34 km=h (Table 2). sition, speed, and TTC, can be derived directly. The trajectories of
Fig. 3. Pictorial view of the case study locations. (Images by Avinash R. Chaudhari.)
both crossing pedestrians and interacting vehicles were extracted Male pedestrians dominated (72%–82%) the pedestrian composi-
for the entire survey duration (Table 2). Fig. 5 represents the meth- tion compared with female pedestrians (18%–28%) at all subject
odology adopted for the extraction of pedestrian–vehicle trajectory. study locations (Fig. 6). Furthermore, middle-aged pedestrians do-
The positions and speeds of the vehicle to compute the time minated the pedestrian composition, followed by young pedestrians
variables that characterize the PRI during the conflict phase were (male/female), and the percentages of elderly and children were the
extracted from the sequence of video frames. Data were extracted lowest (Fig. 7).
when there was a possibility of conflict between the pedestrian Based on types of vehicles observed, it was deemed appropriate
crossing the road and the vehicle approaching the crosswalk. to classify them into six categories: motorized two-wheelers (2W),
The exercise was repeated for stopping as well as passing phases. motorized three-wheelers (3W), car, SUV, bus, and truck. Fig. 8
It is obvious that pedestrian–vehicle interaction will not exist when represents the vehicular traffic compositions recorded at uncon-
there is no pedestrian either on the curb or in the crosswalk. There- trolled midblock crosswalks for all locations. Cars, 2W, and 3W
fore, such conditions were not extracted in this study. Table 3 sum- accounted for a substantial share at all locations (Fig. 8).
marizes the number of samples extracted for the passing, stopping, Furthermore, all possible variations of pedestrian crossing char-
and conflict phases of pedestrian–vehicle interaction for the subject acteristics due to variations in land-use type were incorporated
study locations. (Table 2). In short, the selected study locations incorporated rea-
sonable variation with respect to demographics and pedestrian,
vehicular, and traffic characteristics.
Analysis and Results
Pedestrian–Vehicle Conflict Analysis
Preliminary Analysis
Figs. 6 and 7 characterize pedestrian composition by gender and Validation of Reaction Time
age for the subject study locations. The age variable was catego- The reaction time of vehicle drivers was computed prior to calcu-
rized as children (<15 years), young (15–30 years), middle-aged lating PRI. An appropriate value of reaction time (T r ), 0.7 s, was
(30–50 years), and elderly (>50 years) by visual appearance. chosen as per the maximum deceleration for different categories of
38.34
34.22
24.75
32.07
23.63
23.99
29.61
27.50
32.71
the value of maximum deceleration for the respective vehicle cat-
egories. The data extracted from Location 3 (six-lane divided road)
were used to validate the reaction time. The entry time of the ve-
hicle every 10 and 20 m was taken, from which the vehicle speed
Traffic volume
was computed. The average value of the reaction time of the driver
(PCU=h)
When both the pedestrian and the vehicle move without a change in
980
788
770
818
711
944
510
484
Mixed type
Mixed type
Mixed type
Rail transit
Residential
Rail transit
Shopping
Shopping
Land-use
tors, such as TTC, vehicle speed and PET. Vehicle trajectory slope
was substantially steeper for the eight-lane divided road, resulting
in smaller TTCv. This indicates that approaching vehicles reach
the conflict location with high speed and take less time to reach
Un-marked
Un-marked
Un-marked
Un-marked
Crosswalk
Marked
Marked
Marked
Marked
Marked
the conflict point. On the other hand, the slope of trajectory for the
type
six-lane divided road was moderately steeper than that for the four-
lane divided road, indicating higher TTCv values. Furthermore,
some trajectories indicated a passing phase, indicating that the
Table 2. Details of case study locations with traffic flow data
Two-lane undivided
pedestrian left the conflict point before the vehicle reached the
Eight-lane divided
Six-lane divided
4-lane divided
conflict.
Type of road
ISBT, Chandigarh
V ImpactSpeed for the conflict phase are depicted in Figs. 10(a–f), re-
Bandra, Mumbai
Vakola, Mumbai
Study location/
Varacha, Surat
classification
Fig. 4. Photographs of trajectory data extraction for (a and b) eight-lane divided road; and (c and d) four-lane divided road. (Images by Avinash R.
Chaudhari.)
Female Female
26% 28%
Male
74% Male
72%
Female Female
21% 18%
Male Male
79% 82%
as well. For smaller vehicles (2W, 3W, and cars), the variance in values corresponding to the 15th percentile can be considered as
ΔT and V ImpactSpeed values was higher compared with that for dangerous conflicts, those corresponding to the 50th percentile
heavy vehicles (trucks and buses). can be considered as a potential conflict, and those corresponding
The parameter ΔT (expressed in seconds) is the time difference to the 85th percentile ΔT value can be considered as a normal con-
between T si and TTCvi , and, is a measure to predict conflict prob- flict value. The 15th percentile ΔT values can be considered as the
ability, or it can be considered as a surrogate measure of the prob- lower limit, below which the value is considered as dangerous con-
ability that a crash between a vehicle and a pedestrian occurs in a flict, whereas the 50th percentile can be considered as the limit of
given instance. Previous studies (Davis et al. 2003; Cafiso et al. potential conflict, and the 85th percentile can be considered as safe
2011) were reviewed, and the cumulative distribution curves for crossing (Khanna and Justo 1991).
different locations were plotted to explain the variation with respect The ΔT value (15th percentile value, i.e., dangerous conflict
to governing factors such as the number of lanes (road width), measure) increased with the addition of a lane; it increases by
vehicle type, and pedestrian characteristics (gender and age) 36% from four-lane to six-lane divided roads and by 52% from
[Figs. 11(a–d)]. six-lane to eight-lane divided roads (Table 5). However, the ΔT
The intercepts for the 15th, 50th, and 85th percentiles were ob- value for two-lane divided roads was 70% higher than that for
tained from cumulative frequency plots (Fig. 11) for ΔT (conflict four-lane divided roads. Likewise, the ΔT value (50th percentile
probability) and vehicle speed (m=s) with variation in road geom- value, i.e., conflict measure) also increased with the addition of
etry, vehicle type, and demographic factors (Table 5). The ΔT a lane; it increases by 42.40% from two-lane undivided roads to
four-lane divided roads. It increased by 13% from four-lane to six- undivided roads, which indicates that pedestrians cross wider
lane divided roads and by 80% from six-lane to eight-lane divided streets with greater caution.
roads. For the 85th percentile values (normal conflict), there was The ΔT values also depended on the type of approaching ve-
a marginal increase with the addition of a lane. Furthermore, the hicle. The ΔT values were higher for heavy vehicles and lower for
ΔT value for eight-lane divided roads always was higher than that smaller vehicles such as bikes (2W), cars, and 3W. This may be due
of six-lane divided roads, four-lane divided roads, and two-lane to the aggressive behavior of smaller-vehicle drivers. This shows
that the size of the vehicle affects pedestrian attention. Pedestrians crosswalks on two-lane undivided, four-lane divided, six-lane di-
tended to take the risk to cross the road with rolling behavior, ac- vided, and eight-lane divided roads. The PRI was computed when a
cepting available gaps, if the approaching vehicle was a truck or pedestrian was in the crosswalk and the conflict occurred between
bus. In the case of 2W and 3W, pedestrians crossed with higher pedestrian and approaching vehicle. Student’s t-tests and F-tests
risk, because the ΔT value was much less due to aggressive driving were conducted to verify the statistical differences between the
behavior. average values of PRI obtained for marked crosswalk and un-
Male pedestrians were less conscious about their safety com- marked crosswalks for different locations, based on approaching
pared with females and took higher risks crossing the street with vehicle type. The data first were checked for normality at a 5%
minimum ΔT values. Moreover, child pedestrians crossed the level of significance. The detailed descriptive statistics for PRI
street with a casual approach compared with elderly pedestrians are presented in Tables 7 and 8.
with smaller ΔT values. Furthermore, the average vehicle speed Statistical analysis (Tables 7 and 8) showed no significant differ-
increased with the addition of a lane; it increased by 8.57% from ence in PRI values between marked and unmarked crosswalks for
four-lane to six-lane divided roads, and by 56.82% from six-lane to 3W and bus/truck, except bus/truck on eight-lane divided roads,
eight-lane divided roads. However, in the case of two-lane undi- which may be due to high speeds. There was a significant differ-
vided roads, the traffic flow was low with higher headway. In ad- ence in the PRI between marked and unmarked crosswalks for
dition, pedestrian flow was low due to high average approaching smaller aggressive vehicle types, i.e., 2W and cars. The average
speed (Table 2). Furthermore, the speed of 2W was higher than that value of PRI also was compared for statistical significance based
of 3W, cars, and heavy vehicles (trucks and buses) by 16.66%, on variation in the number of lanes (two-lane undivided and four-
7.58%, and 23.13%, respectively. In addition, male pedestrians lane, six-lane, and eight-lane divided roads) for different vehicle
took greater risk while crossing the road than did female pedes- types (Table 9).
trians. Furthermore, the drivers of approaching vehicle reduced From Table 9, it can be inferred that the PRI also depended on
the speed of the vehicle by 5.03% when a female pedestrian the type of vehicle. PRI value for 2W was significantly higher than
was on the crosswalk compared with a male pedestrian. The speed that for cars and 3W. The PRI values for heavy vehicles (truck/bus)
of the approaching vehicle was 25.70% higher when young and were smaller than those for 2W, 3W, and cars. The PRI value for the
middle-aged pedestrian were present and 37.71% lower when an
smaller vehicles was higher than that for heavy vehicles. This may
elderly pedestrian was present on the crosswalk, compared with
be due to the aggressive behavior of the drivers, who approached
the situation for child pedestrians.
the crosswalk at high speed. Furthermore, the PRI values increased
Computation of PRI by 37.35% with the addition of lanes. The PRI value for eight-lane
The proposed methodology for computing PRI was applied to the divided roads was higher than that for two-lane undivided and
subject study locations. The mean vehicle speed over the vehicle four-lane and six-lane divided roads, which may be due to the
trajectory within the marked crosswalk area and a border equal higher average approaching speed of the vehicles. On average,
to the width of the opening for the unmarked crosswalk was used 4%–12% of drivers reduced their vehicular speed with safe stop-
as the vehicle approaching speed for each vehicle for different ping distance to allow pedestrians to cross the road, whereas
crosswalk locations. First, the vehicle trajectory of vehicles and pe- 88%–96% of drivers did not allow pedestrians to cross; instead,
destrians in the conflict area on four-lane divided, six-lane divided, they increased the approaching vehicle speed at the crosswalk to
and eight-lane divided roads are plotted in Figs. 9(a–c) as exam- avoid delay at the midblock. However, this specific behavior in-
ples. The intersection of the trajectories of both elements is iden- creases crash risk. Furthermore, the traffic volume on eight-lane
tified as a conflict. At the conflict point, the speed of approaching divided roads was higher than that on six-lane and four-lane di-
vehicle (VI) and the time to reach the conflict point (TTcv ) were vided and two-lane undivided roads (Table 2), which implies that
determined from each trajectory of vehicles and pedestrians. The as the traffic volume increases, the PRI value also increases. The
related parameter was computed using these trajectories for a se- PRI increased (37.35%) with the addition of every lane. The find-
lected sample of vehicles on four-lane, six-lane, and eight-lane di- ings also indicate that the PRI value for eight-lane divided roads
vided roads using these values. The computed PRIs are presented in was higher than that for four-lane and six-lane divided roads. This
Table 6. was due to the higher average approaching speed of vehicles.
In the same way, the traffic conflicts were detected, and the Furthermore, to determine the effect of approaching vehicle
corresponding PRI values were estimated for all the locations to speed, traffic volume, and pedestrian volume crossing the road on
evaluate the prevailing safety level, for unmarked and marked PRI, different plots were generated depicting possible relationships
0
(a)
(b)
(c)
Fig. 9. Vehicle–pedestrian trajectory plots for (a) four-lane divided road; (b) six-lane divided road; and (c) eight-lane divided road.
(a)
(b)
(c)
Fig. 10. V ImpactSpeed versus ΔT by vehicle type during different phases: (a) passing phase for smaller vehicle (2W and 3W); (b) passing phase for
smaller vehicle (car) and heavy vehicle (bus/truck); (c) stopping phase for smaller vehicle (2W and 3W); (d) stopping phase for smaller vehicle (car)
and heavy vehicle (bus/truck); (e) conflict phase for smaller vehicle (2W and 3W); and (f) conflict phase for smaller vehicle (car) and heavy vehicle
(bus/truck).
(d)
(e)
(f)
Fig. 10. (Continued.)
PRI value. Hence, higher approaching vehicle speed, traffic volume, The PET values of different vehicles types when compared for
and pedestrian volume intensify pedestrian–vehicle interaction, similar geometry were lower for 2W, which indicates that conflicts
which is a complex phenomenon. Nevertheless, PRI as a measure with 2W are more critical compared with other vehicle categories
of safety can serve practitioners as well as researchers by develop- (Table 10). Similarly, PRI values for conflicts with 2W were higher
ing pedestrian crossing warrants to monitor the safety levels and than those with other vehicles, which indicates that conflicts
operational efficiency at midblock crosswalks. with 2W are more critical than those with other vehicle categories.
Furthermore, average PET values increased with a change in
Comparison with Other Traffic Conflict Techniques road geometry. For instance, higher PET values were found for
The results of the proposed methodology were compared with other two-lane divided road geometry, followed by four-lane, six-lane
TCTs, namely PET, to corroborate the developed methodology for and eight-lane divided road geometry. This indicates that the
assessing pedestrian safety. It is obvious that smaller values of PET severity of conflicts is lesser for two-lane undivided road geom-
indicate greater severity of the conflict. On the other hand, higher etry compared to other road configurations. This can be attributed
values of PRI indicate higher risk and severity of the conflict. A to the presence of contraflow. Higher PRI values were found
comparative summary is presented in Table 10. for eight-lane divided road geometry, followed by six-lane and
(a)
(b)
(c)
Fig. 11. Cumulative frequency plots for ΔT (conflict probability) and vehicle speed by (a) number of lanes; (b) vehicle type; (c) gender; and (d) age.
(d)
Table 5. Variation of ΔT, vehicle speed based on road, vehicle type, and demographic factors
ΔT (s) Vehicle speed (m=s)
No. Classification Particular 15th 50th 85th 15th 50th 85th
01 Lane Two -lane undivided 2.25 3.676 5.11 20.6 29.40 38.20
Four- lane divided 0.68 2.727 4.86 8.40 21.80 35.20
Six-lane divided 1.06 3.083 5.11 13.1 23.67 34.30
Eight-lane divided 2.230 5.535 8.83 9.70 37.12 64.50
02 Vehicle 2W 0.76 3.094 5.13 15.88 28.75 34.00
3W 0.69 3.529 6.30 11.56 23.95 36.35
Car 0.79 3.015 5.38 11.23 26.57 41.91
Bus/truck 1.06 4.488 8.02 10.19 22.10 41.62
03 Gender Male 0.75 3.52 6.29 12.90 26.42 39.90
Female 1.08 3.90 6.71 9.30 25.09 42.70
04 Age Child 0.61 3.239 5.87 9.60 21.05 32.50
Young 0.87 3.816 6.77 11.20 26.46 40.50
Middle-aged 0.99 3.758 6.52 13.00 26.79 41.70
Elderly 1.36 4.315 7.25 12.40 19.53 26.70
higher compared with that for six-lane and four lane divided road- through the origin (zero) (Fig. 13). The R2 value of the validation
ways. Pedestrian volume, vehicular speeds, and vehicular volume was 0.728. The results showed that the developed multilinear re-
were considered as independent variables to model PRI. Three dif- gression model has good prediction proficiency for estimating
ferent models were developed. The first model considers only PRI the PRI value for road crossing behavior of pedestrians at uncon-
as a function of vehicular speed; the second model considers PRI as trolled midblock crosswalk sections in Indian conditions. Mean
a function of vehicular speed and vehicular volume; and the third absolute percentage error (MAPE) was 5.54%. According to the
model considers PRI as a function of vehicular speed, vehicle vol- Lewis scale of interpretation, estimation of accuracy for any fore-
ume, and pedestrian volume. The models were developed with 60% cast with a MAPE value of less than 10% can be considered rea-
of the samples and were validated using the remaining 40% of sam- sonably accurate, 11%–20% can be considered good, 21%–50%
ples. The models are summarized in Table 12. Model 3. i.e., PRI as can be considered reasonable, and 51% or more can be considered
a function of vehicular speed, vehicle volume, and pedestrian vol- inaccurate (Kenneth and Ronald 1982). Therefore, the calibrated
ume, is considered as final model based on the R2 values. PRI model can be deemed accurate for predicting PRI values
The calibrated PRI model was validated with 40% of the re- using a set of vehicular speeds and pedestrian and vehicular
maining data. The predicted values were calculated by substituting volumes. Other variables, such as the presence of marked or un-
the values of variables in the obtained model. Thereafter, the pre- marked crosswalks, land-use type, and road geometry, were con-
dicted PRI values were compared with the observed PRI values. sidered for modeling. However, these variables were found to be
Observed and predicted values were plotted with a line passing insignificant.
(a) (b)
(c)
Fig. 12. PRI versus (a) vehicle speed; (b) traffic volume; and (c) pedestrian volume.
Davis, G., K. Sanderson, and S. Davuluri. 2003. “A vehicle-pedestrian complex intersections.” Transp. Res. Rec. 2264 (1): 148–155. https://
collision model for neighborhood traffic control.” In Proc., 82nd Trans- doi.org/10.3141/2264-17.
portation Research Board Annual Meeting. Washington, DC: Transpor- Sayed, T., M. H. Zaki, and J. Autey. 2013. “A novel approach for diag-
tation Research Board. nosing road safety issues using automated computer vision techniques.”
Gettmann, D., L. Pu, T. Sayed, and S. Shelby. 2008. Surrogate safety In Proc., 16th Road Safety on Four Continents Conf., 15–17. Beijing:
assessment model and validation: Final report. Rep. No. FHWA- Road Safety on Four Continents.
HRT-08-051. Washington, DC: Federal Highway Administration. Silcock, D. T., R. Walker, and T. Selby. 1998. “Pedestrians at risk.”
Hannawald, L., and F. Kauer. 2004. Equal effectiveness study on pedestrian In Proc., European Transport Conf., 209–220. London: PTRC Educa-
protection. Dresden, Germany: Technische Univ. tion and Research Services Limited.
Hayward, J. C. 1972. Near-miss determination through use of a scale of St-Aubin, P., N. Saunier, L. F. Miranda-Moreno, and K. Ismail. 2013. “Use
danger. Rep. No. TTSC 7115. State College, PA: Pennsylvania State of computer vision data for detailed driver behaviour analysis and
Univ. trajectory interpretation at roundabouts.” Transp. Res. Board 2389 (1):
Jiang, X., W. Wang, K. Bengler, and W. Guo. 2015. “Analyses of pedestrian 65–77. https://doi.org/10.3141/2389-07.
behavior on mid-block unsignalized crosswalk comparing Chinese and Tarko, A., and G. Davis. 2009. “Surrogate measures of safety.” In Proc.,
German cases.” Adv. Mech. Eng. 7 (11): 168781401561046. https://doi 88th Transportation Research Board Annual Meeting. Washington,
.org/10.1177/1687814015610468. DC: Transportation Research Board.
Jones, D. K., K. R. Evenson, D. A. Rodriguez, and S. A. Aytur. 2010. Toran Pour, A., S. Moridpour, R. Tay, and A. Rajabifard. 2017. “Modelling
“Addressing pedestrian safety: A content analysis of pedestrian master pedestrian crash severity at mid-blocks.” Transportmetrica A: Transp.
plans in North Carolina.” Traffic Inj. Prev. 11 (1): 57–65. https://doi.org Sci. 13 (3): 273–297. https://doi.org/10.1080/23249935.2016.1256355.
/10.1080/15389580903434199. Turner, S., K. Fitzpatrick, M. Brewer, and E. S. Park. 2006. “Motorist
Kenneth, D. L., and K. K. Ronald. 1982. Advances in business and man- yielding to pedestrians at unsignalized intersections: Findings from a
agement forecasting. Bingley, UK: Emerald Books. national study on improving pedestrian safety.” Transp. Res. Rec.
Khanna, S. K., and C. E. G. Justo. 1991. Highway engineering. Roorkee, 1982 (1): 1–12. https://doi.org/10.1177/0361198106198200102.
India: Nem Chand & Bros. van der Horst A. R. A., and J. H. Kraay. 1986. “The Dutch conflict obser-
Laureshyn, A., A. Svensson, and C. Hydén. 2010. “Evaluation of traffic vation technique ‘DOCTOR’.” In Proc., ICTCT Workshop. Budapest,
safety based on micro-level behavioural data: Theoretical framework Hungary: KTI Institute for Transport Sciences.
and first implementation.” Accid. Anal. Prev. 42 (6): 1637–1646. Varhelyi, A. 1996. Dynamic speed adaptation based on information
https://doi.org/10.1016/j.aap.2010.03.021. technology—A theoretical background. Bulletin 142. Lund, Sweden:
Maurya, A. K., and P. S. Bokare. 2012. “Study of deceleration behaviour of Lund Univ.
different vehicle types.” Int. J. Traffic Transp. Eng. 2 (3): 253–270. Vaughn, R. 1997. Avoiding the emerging pedestrian: A mathematical model.
https://doi.org/10.7708/ijtte.2012.2(3).07. SAE Technical Paper 970962. Warrendale, PA: SAE International.
Migletz, D. J., W. D. Glauz, and K. M. Bauer. 1985. Relationships Venthuruthiyil, S. P., and C. Mallikarjuna. 2018. “Trajectory reconstruction
between traffic conflicts and accidents. Rep. No. FHWA/RD-84/042. using locally weighted regression: A new methodology to identify the
Washington, DC: USDOT, Federal Highway Administration. optimum window size and polynomial order.” Transportmetrica A:
MoRTH (Ministry of Road Transport and Highways). 2013. Study on road Transp. Sci. 14 (10): 881–900. https://doi.org/10.1080/23249935.2018
accidents in India. New Delhi, India: MoRTH. .1449032.
NCSA (National Center for Statistics and Analysis). 2013. Pedestrians: Wu, J., E. Radwan, and H. Abou-Senna. 2018. “Assessment of pedestrian-
Traffic safety facts, 2011 data. Rep. No. DOT HS 811 748. Washington, vehicle conflicts with different potential risk factors at midblock
DC: National Highway Traffic Safety Administration. crossings based on driving simulator experiment.” Adv. Transp. Stud.
Nhan, C., L. Rothman, M. Slater, and A. Howard. 2009. “Back-over 44 (Apr): 33–46.
collisions in child pedestrians from the Canadian Hospitals Injury Re- Yao, J., J. Yang, and D. Otte. 2007. “Head injuries in child pedestrian
porting and Prevention Program.” Traffic Inj. Prev. 10 (4): 350–353. accidents in depth case analysis and reconstructions.” Traffic Inj. Prev.
https://doi.org/10.1080/15389580902995166. 8 (1): 94–100. https://doi.org/10.1080/15389580600944243.
NHTSA (National Highway Traffic Safety Administration). 2015. Zegeer, C., J. R. Stewart, H. H. Huang, and P. A. Lagerwey. 2001. Safety
“Pedestrian safety enforcement operations: A how-to guide.” Accessed effects of marked vs. unmarked crosswalks at uncontrolled locations:
October 31, 2017. http://www.nhtsa.gov/staticfiles/nti/pdf/812059 Executive summary and recommended guidelines. Report No. FHWA-
-PedestrianSafetyEnforceOperaHowToGuide.pdf. RD-01-075. Washington, DC: Federal Highway Administration.