Professional Documents
Culture Documents
Safety Science
journal homepage: www.elsevier.com/locate/safety
A R T I C L E I N F O A B S T R A C T
Keywords: As a consequence of the rapid growth of vehicular traffic, there is an increase in interactions between vehicles
Threshold values and pedestrians. The severity of these interactions varies with pedestrian, vehicle and roadway geometric
Risk indicator characteristics. In the absence of real crash data, Surrogate Safety Measures (SSMs) are used to analyse the
Surrogate safety measures
pedestrian-vehicle (P-V) interactions. The present study is intended to propose threshold risk indicator (RI)
Pedestrian-vehicle interaction
Post encroachment time
values for severe P-V interactions using both pedestrian and vehicle characteristics. A multilinear regression
Support vector machines (MLR) P-V interaction model was developed using SPSS (Statistical Package for the Social Sciences) software.
Videography method was used to collect traffic data from two 4-legged uncontrolled intersections. Pedestrian
and vehicular data were extracted from the video using DataFromSky viewer software and risk indicator was
calculated using post encroachment time and approaching vehicular speed. The interactions between pedestrians
and vehicles were classified as normal conflicts and severe conflicts based on visual observations during the data
extraction process. Python interface with support vector machines (SVM) algorithm was used to get threshold RI
values for various pedestrian (gender and speed) and vehicle (type) characteristics.
From SVM results, it was observed that the threshold RI value for severe interactions decreases as the
pedestrian crossing speed increases for the same vehicle and pedestrian characteristics. MLR results showed that
pedestrian gender, age and speed, vehicle type and speed, interaction location and crossing position have a
significant effect on RI. The results can be used to evaluate pedestrian-vehicle interaction severity level at an
uncontrolled intersection.
* Corresponding author.
E-mail addresses: lalamgovinda2246@student.nitw.ac.in (L. Govinda), ravikvr@nitw.ac.in (K.V.R. Ravi Shankar).
https://doi.org/10.1016/j.ssci.2022.105806
Received 9 March 2021; Received in revised form 22 April 2022; Accepted 26 April 2022
Available online 2 May 2022
0925-7535/© 2022 Elsevier Ltd. All rights reserved.
L. Govinda et al. Safety Science 153 (2022) 105806
risk of a potential collision, a novel surrogate safety indicator i.e. risk different vehicle types based on the cumulative frequency distribution.
indicator was used involving approaching vehicle speeds and post There is no chance of interaction between pedestrian and vehicle when
encroachment time (Scholl et al., 2019). Risk indicator (RI) is defined as PET value is greater than 5.5 s and highly dangerous conflict is possible
the ratio of approaching vehicle speed over post encroachment time. when the PET value is less than or equal to 2.0 sec. The threshold PET
value for highly dangerous conflict was 2.5 s, 3.06 s, 3.83 s, 2.37 s, and
Approaching Vehicle Speed
Risk Indicator = (1) 2.25 s for car, two-wheeler, LCV (light commercial vehicle), HCV (heavy
Post Encroachment Time
commercial vehicle), and auto respectively.
Most of the previous studies have focused on the analysis and
2. Background
modelling of P-V interactions but very few studies have proposed
threshold values of SSMs to classify the P-V interactions severity levels.
Conducting traffic conflict studies could be an effective alternative
Some studies classified interactions based on vehicular speeds (Scholl
for safety analysis because of the high correlation between the historical
et al., 2019) while other studies classified interactions using pedestrian
crash data and the observed traffic conflicts (Qi and Yuan, 2012). A large
gender, speed and vehicle type (Marisamynathan and Vedagiri, 2020).
number of studies on pedestrian safety and risk behaviour identified
These studies did not include both pedestrian and vehicular speeds at a
factors which influence pedestrian safety. Previous studies identified
time to classify P-V interaction severity levels. The present study was
various pedestrian, vehicle, road, and environmental factors that influ
intended to develop the threshold values of SSMs using both pedestrian
ence pedestrian safety and risk-taking behaviour while crossing the
(gender and speed) and vehicle (type and speed) characteristics. Also, it
road. Pedestrian age has the most significant effect while gender has a
intended to develop a multilinear P-V interaction model.
small effect on pedestrian walking speeds (Gates et al., 2006; Rav
ishankar and Nair, 2018). Female and older pedestrians take lower risk
compared to other pedestrians (Harrell, 1991; Jain et al., 2014; Ols 3. Data collection and extraction
zewski et al., 2015). Around 80% of fatal accidents occur in the case of
elderly pedestrians, drunk pedestrians, and pedestrians in hours of The present study was conducted at two 4-legged uncontrolled in
darkness (Jensen, 1999). Pedestrians involved in past traffic accidents tersections in Warangal and Nagpur cities in India. These two locations
take lower risk and the risk is low for the pedestrian near to the curb were selected because of the rapid growth in population and traffic over
rather than the one close to the central refuse island (Hamed, 2001). The the past two decades. Mixed land use characteristics were observed at
risk associated with the presence of median would be more compared to both study locations. The video-based data collection system was more
that without median (Jimenez-Mejias et al., 2016; Olszewski et al., appropriate to analyse P-V interactions and dilemma zone. Two high-
2015) and the risk behaviour for right-turning vehicles was relatively resolution cameras were fixed at an elevation in such a way that both
high compared to other vehicles (Chen et al., 2017). The risk of injury pedestrian movement and vehicle movement could be captured. 4 h
was significantly lower in case of a pedestrian crossing at zebra crossings (morning and evening time during peak periods) traffic data was
than one crossing at other places (Keall, 1995). Number of lanes, the collected from each study location in February 2019. Video recording
proportion of young male pedestrians and land use patterns had a sig mainly focused on the movement of vehicles and pedestrian road
nificant effect on the severity of P-V interactions (Sandt and Zegeer, crossing behaviour. The geometric details of selected study locations
2006). Width of opposing direction, opposing through vehicle volume, measured from the field and the location view are shown in Fig. 1 below.
left turn vehicle volume, and pedestrian volume were the significant Warangal intersection has a two-lane divided road with 3.5 m lane width
factors that affect pedestrian safety (Qi and Yuan, 2012). along the major road and minor roads. Nagpur intersection has a two-
In the absence of real crash data, surrogate safety measures (SSMs) lane divided road along major road and a two-lane undivided road
were used to analyse P-V interactions using videography data. Most along the minor road with 3.5 m lane width. The width of the median at
commonly used SSMs to analyse P-V interactions are Time To Collision both intersections was observed to be 1.0 m.
(TTC), Post Encroachment Time (PET), Time Difference To Collision DataFromSky viewer software was used to extract pedestrian and
(TDTC), and Gap time (GT) etc. Most of the previous studies analysed P- vehicle parameters from the video. It can be used to extract parameters
V interactions using SSMs but very few studies have proposed threshold like speed, acceleration/deceleration and trajectories of both vehicles
values of SSMs for different P-V interaction severity levels. The fre and pedestrians, PET, traffic volume counts, vehicle type, and travel
quency of conflicts within the crosswalk (PET < 3 s) was much more distance etc. High accuracy of data was obtained with this software and
than those of outside crosswalk indicating less risk of crossing within the the data extraction process was also quite easy. A screenshot of Data
crosswalk (Chen et al., 2017). T-intersections were more dangerous than FromSky viewer software is shown in Fig. 2.
Cross intersections due to high conflict rate and there was a possibility of The pedestrian age is classified based on the visual observation
conflict between pedestrian and vehicle when TTC < 1.5 s and PET < during the data extraction process. The pedestrian age is classified as
3.0 s (Lord, 1996). (Almodfer et al., 2016) developed a novel indicator child pedestrians (<=15 years), young pedestrians (16–30 years),
(Lane based Post Encroachment Time (LPET)) to evaluate pedestrian- middle age pedestrians (31–60) years and old pedestrians (>60 years)
vehicle conflict analysis. It was observed that pedestrian-vehicle lane- (Patra et al., 2017). The interactions were classified as normal or severe
based conflicts were not evenly distributed and far lanes contributed to based on the visual observation during the data extraction process and
serious conflicts than nearer lanes. The number of severe conflicts de the definitions of severe and normal interactions for visual classification
creases as the waiting time of pedestrian increases and the severity of wareere given below. 2954P-V interactions were extracted from the
conflicts was not affected by the lane position and the stages of the video and the compositions of pedestrians and vehicles are shown in
crossing. Fig. 3 below. 47% of total P-V interactions were observed in the case of
(Scholl et al., 2019) proposed a novel surrogate safety measure i.e. two-wheelers (2W) and the percentage of interactions with
risk index or risk indicator (RI) using post encroachment time and three-wheelers (3W) and cars were 26% and 22% respectively. LCVs and
approaching vehicle speed and the proposed threshold ln(RI) values for HCVs were not considered in the analysis due to a limited number of P-V
high, moderate, and low-risk interactions were 3.97, 2.94, and 2.26 interactions (3% of LCVs and 2% HCVs). 56% of extracted samples were
respectively. (Fu et al., 2016) classified PET < 1.5 s as dangerous con related to middle age pedestrians and only 1% of extracted samples were
flicts and found that these conflicts were higher at night time compared related to child pedestrian interactions.
to day. Pedestrian safety is lower at night compared to day. (Mar Severe conflict: Both or one of the two road users must stop or
isamynathan and Vedagiri, 2020) developed threshold PET values for change their speeds to avoid the accident. The possibility of an accident
highly dangerous conflicts, dangerous conflicts, and no conflicts for is high in this situation.
Normal conflict: Both pedestrian and vehicle move at their normal
2
L. Govinda et al. Safety Science 153 (2022) 105806
Fig. 3. (a) Vehicle composition (b) Pedestrian composition with respect to age.
3
L. Govinda et al. Safety Science 153 (2022) 105806
Fig. 4. Pedestrian crossing speeds with respect to (a) pedestrian gender (b) pedestrian age.
4
L. Govinda et al. Safety Science 153 (2022) 105806
Fig. 6. Average RI of P-V interaction with respect to (a) pedestrian age (b) pedestrian gender.
Fig. 7. Average RI of P-V interaction with respect to (a) vehicle type (b) location of P-V interactions.
5
L. Govinda et al. Safety Science 153 (2022) 105806
Fig. 9. Classification of P-V interactions for (a) male- 3W (b) female- 3W.
Fig. 10. Classification of P-V interactions for (a) male-car (b) female- car.
6
L. Govinda et al. Safety Science 153 (2022) 105806
Table 1
Threshold values of risk indicator for severe interactions based on pedestrian and vehicle attributes.
Category of P-V interaction Pedestrian crossing speeds (m/s)
Thresholds for risk indicator (RI) M-2W 8.8–8.4 8.4–8.0 8.0–7.7 7.7–7.4 7.4–7.1 7.1–6.8
M-3W 7.5–7.0 7.0–6.7 6.7–6.4 6.4–6.0 6.0–5.6 5.6–5.1
M-Car 5.8–5.5 5.5–5.2 5.2–4.8 4.8–4.4 4.4–3.9 3.9–3.4
F-2W 7.9–7.4 7.4–7.0 7.0–6.6 6.6–6.2 6.2–5.8 5.8–5.4
F-3W 7.3–7.0 7.0–6.7 6.7–6.5 6.5–6.2 6.2–6.0 6.0–5.7
F-Car 6.0–5.7 5.7–5.4 5.4–5.1 5.1–4.8 4.8–4.5 4.5–4.3
Table 2
Confusion matrices for testing data based on SVM.
variables with p < 0.05 were included in the model. Pedestrian age (p- development of any model. Validation shows the realistic representation
value 0.007), gender (p-value 0.002), speed (p-value 0.012), vehicle of the actual system. For the present study, 25% of data was used for
type (p-value 0.003), lane position of P-V interaction (p-value 0.001), validation of the proposed MLR model. A graph was plotted between
interaction location (p-value 0.007), and crossing type (p-value 0.001) observed RI values on the x-axis and predicted/expected RI values y-axis
have significant effect with the risk indicator. The mode developed after and shown in Fig. 11. The closeness of the data points to the 45-degree
considering 6 independent variables is shown in equation (3). line indicates the validation of the model. A good level of confidence (R2
= 0.637) was observed between the observed and expected RI values.
RI = 12.99 − 0.722 × Gen − 0.850 × Age − 1.062 × PS − 0.352
× VT − 0.361 × IL − 1.277 × CP (3) 6. Conclusions
All the six independent variables (gender, age, PS, VT, IL, and CP)
Most of the previous studies (Almodfer et al., 2016; Chen et al., 2017;
show a negative correlation with the risk indicator. The severity of P-V
Fu et al., 2016; Lord, 1996; Scholl et al., 2019) analysed P-V interactions
interactions decreases with a decrease in RI value. The value of RI de
using SSMs but very few studies (Marisamynathan and Vedagiri, 2020;
creases with increase in pedestrian age and gender. Also, it decreases
Shah and Vedagiri, 2017) have proposed threshold values for P-V
when the interaction occurs between male pedestrian and two-wheeler.
interaction severity levels using either pedestrian or vehicle character
The severity levels of P-V interactions are higher when an interaction
istics. In the present study, the severity level of P-V interactions was
occurs at near side of pedestrians and entry point to vehicles (Near_
classified based on visual observation during data extraction from video
Entry) compared to the far side of pedestrians and exit to vehicles
and proposed threshold values for severe P-V interactions using both
(Far_Exit). Higher approaching vehicular speeds at the entry point
pedestrian and vehicle characteristics. Videography data was collected
compared to the exit point is the reason for higher levels of severity.
from two 4-legged uncontrolled intersections and the required pedes
Severity levels are lower when the pedestrians cross the road at cross
trian and vehicle data was extracted from video using DataFromSky
walk compared to crossing outside crosswalk. Also, it was observed that
software.
the severity levels are higher when the conflicting vehicle is 2W
Python interface with SVM coding was used to estimate the threshold
compared to other vehicle types.
values of RI for different combinations of pedestrian gender, speed and
Calibration and validation are two important aspects for the
vehicle type. The severity level of P-V interaction inversely correlated
7
L. Govinda et al. Safety Science 153 (2022) 105806
Table 3 were observed due to the flexibility of pedestrians to change their speeds
Dependent and independent variables for the MLR model. with respect to vehicle speed and position. The influence of speed and
S. Variable Description Category/Units/Code vehicle size on pedestrian perception for risk-taking behaviour is the
No reason for higher severity levels in case of two-wheelers. The perfor
1 Risk indicator The ratio of approaching Dependent variable mance of classified data was described using confusion matrix and ac
(RI) vehicular speed to the post curacy was found to be more than 80% in all cases (from Table 2).
encroachment time SPSS 20 software was used to carry out multilinear regression anal
2 Pedestrian Based on visual appearance Male-1, Female-2 ysis at 95% confidence interval and the variables with t-value more than
gender (Gen) classified as Male and Female
3 Pedestrian age Based on visual appearance Child-1, Young age-2,
t-critical (1.895) were included in the model. From the MLR model, it
(Age) classified as Child (<10 years) Middle age-3, Old was concluded that the risk indicator mostly depends on pedestrian age,
Young (10–25 years), Middle age-4 gender, crossing speed, crossing type, vehicle type, lane position, and
(20–60 years) and Old (>60 interaction location. The proposed model can be used to estimate the
years)
value of RI for different pedestrian and vehicle characteristics. The
4 Pedestrian Crossing speeds of pedestrians Ped speed(m/s)
speed (PS) severity of P-V interaction is classified by comparing the model results
5 Vehicle type Classified based on visual Two-wheeler-1, with the table value (threshold RI values proposed in Table 1). The
(VT) appearance as Two-wheelers Three-wheeler-2, Car- proposed model was validated on a 45-degree line and it showed a good
(2w), Three wheelers (3w), Car, 3, LCV-4, HCV-5 level of confidence (R2 = 0.637) between the observed and the expected
Light Commercial Vehicle (LCV)
or Bus
RI values.
6 Interaction Whether the pedestrian-vehicle Near_Entry-1, The present study was limited to estimating the threshold RI values
location (IL) interaction occurs at entry or exit Near_Exit-2, for two-wheelers, three-wheelers and cars due to fewer number of LCVs
of the intersection Far_Entry-3, Far_Exit- and HCVs. The inclusion of threshold values for these two vehicle types
4
would give better information towards P-V interaction analysis. The
7 Crossing Whether the pedestrian-vehicle Within-1, Away-2
position(CP) interaction occurs within the present study classified the threshold values based on pedestrian gender,
crosswalk or away from the speed and vehicle type. In future, it can be extended to define the
crosswalk threshold values of SSMs based on the pedestrian age as well.
References
Allen, B.L., Shin, B.T., 1967. Analysis of Traffic Conflicts and Collisions. Transp. Res. Rec.
67–74.
Almodfer, R., Xiong, S., Fang, Z., Kong, X., Zheng, S., 2016. Quantitative analysis of lane-
based pedestrian-vehicle conflict at a non-signalized marked crosswalk. Transport.
Res. Part F Traffic Psychol. Behav. 42, 468–478. https://doi.org/10.1016/j.
trf.2015.07.004.
Chen, P., Zeng, W., Yu, G., Wang, Y., 2017. Surrogate Safety Analysis of Pedestrian-
Vehicle Conflict at Intersections Using Unmanned Aerial Vehicle Videos. J. Adv.
Transport. 2017, 1–12. https://doi.org/10.1155/2017/5202150.
Fu, T., Miranda-Moreno, L., Saunier, N., 2016. Pedestrian crosswalk safety at
nonsignalized crossings during nighttime: Use of thermal video data and surrogate
safety measures. Transport. Res. Rec.: J. Transport. Res. Board 2586 (1), 90–99.
https://doi.org/10.3141/2586-10.
Gates, B.T.J., Noyce, D.A., Bill, A.R., Van Ee, N., Gates, T.J., 2006. Recommended
Fig. 11. Validation of the MLR model. Walking Speeds for Pedestrian Clearance Timing Based on Pedestrian
Characteristics. Transport. Res. Rec.: J. Transport. Res. Board 1982, 38–47.
Govinda, L., Abhigna, D., Nair, P.M., Ravishankar, K.V.R., 2020. Comparative Study of
with PET and directly correlated with risk indicator. The severity level Pedestrian Crossing Behaviour at Uncontrolled Intersection and Midblock Locations.
was higher when male pedestrians were involved in the interactions due Transp. Res. Procedia 48, 698–706. https://doi.org/10.1016/j.trpro.2020.08.070.
Hamed, M.M., 2001. Analysis of pedestrians’ behavior at pedestrian crossings. Saf. Sci.
to higher risk-taking behaviour. The severity level decreases with in 38 (1), 63–82. https://doi.org/10.1016/S0925-7535(00)00058-8.
crease in pedestrian crossing speed for the same pedestrian gender and
vehicle type. Lower interaction severity levels at higher crossing speeds
8
L. Govinda et al. Safety Science 153 (2022) 105806
Harrell, W.A., 1991. Factors influencing pedestrian cautiousness in crossing streets. Olszewski, P., Szagała, P., Wolański, M., Zielińska, A., 2015. Pedestrian fatality risk in
J. Soc. Psychol. 131 (3), 367–372. https://doi.org/10.1080/ accidents at unsignalized zebra crosswalks in Poland. Accid. Anal. Prev. 84, 83–91.
00224545.1991.9713863. https://doi.org/10.1016/j.aap.2015.08.008.
Jain, A., Gupta, A., Rastogi, R., 2014. Pedestrian crossing behaviour analysis at Patra, M, Sala, E., Ravishankar, K.V.R., 2017. Evaluation of pedestrian flow
intersections. Int. J. Traffic Transp. Eng. 4 (1), 103–116. https://doi.org/10.7708/ characteristics across different facilities inside a railway station. Transp. Res.
ijtte.2014.4(1).08. Procedia 25, 4763–4770. https://doi.org/10.1016/j.trpro.2017.05.488.
Jensen, S., 1999. Pedestrian safety in Denmark. Transp. Res. Rec. 1674 (1), 61–69. Qi, Y.i., Yuan, P., 2012. Pedestrian safety at intersections under control of permissive
https://doi.org/10.3141/1674-09. left-turn signal. Transp. Res. Rec. 2299 (1), 91–99. https://doi.org/10.3141/2299-
Jimenez-Mejias, E., Martinez-Ruiz, V., Amezcua-Prieto, C., Olmedo-Requena, R., Luna- 10.
Del-Castillo, J.D.D., Lardelli-Claret, P., 2016. Pedestrian- and driver-related factors Ravishankar, K.V.R., Nair, P.M., 2018. Pedestrian risk analysis at uncontrolled midblock
associated with the risk of causing collisions involving pedestrians in Spain. Accid. and unsignalised intersections. J. Traffic Transport. Eng. (Engl. Ed.) 5 (2), 137–147.
Anal. Prev. 92, 211–218. https://doi.org/10.1016/j.aap.2016.03.021. https://doi.org/10.1016/j.jtte.2017.06.005.
Keall, M.D., 1995. Pedestrian exposure to risk of road accident in New Zealand. Accid. Sandt, L., Zegeer, C.V., 2006. Characteristics related to midblock pedestrian-vehicle
Anal. Prev. 27 (5), 729–740. https://doi.org/10.1016/0001-4575(95)00019-V. crashes and potential treatments. Transport. Res. Rec.: J. Transport. Res. Board
Lobjois, R., Benguigui, N., Cavallo, V., 2013. The effects of age and traffic density on 1982, 113–121. https://doi.org/10.3141/1982-16.
street-crossing behavior. Accid. Anal. Prev. 53, 166–175. https://doi.org/10.1016/j. Scholl, L., Elagaty, M., Ledezma-Navarro, B., Zamora, E., Miranda-Moreno, L., 2019.
aap.2012.12.028. A surrogate video-based safety methodology for diagnosis and evaluation of low-cost
Lord, D., 1996. Analysis of pedestrian conflicts with left-turning traffic. Transp. Res. Rec. pedestrian-safety countermeasures: The case of Cochabamba, Bolivia. Sustainability
1538, 61–67. https://doi.org/10.3141/1538-08. 11, 1–22. https://doi.org/10.3390/su11174737.
Marisamynathan, S., Vedagiri, P., 2020. Pedestrian Safety Evaluation of Signalized Shah, H., Vedagiri, P., 2017. Evaluation of Pedestrian Safety at Unsignalised Intersection
Intersections using Surrogate Safety Measures. Transport 35, 48–56. https://doi.org/ under Mix Traffic Condition Using Surrogate Safety Measures. University of South
10.3846/transport.2020.12157. Australia, Adeliade.
MORTH-2018, 2019. Ministry of Road Transport and Highways (MORTH)(2018)-Road W. H. Organization, 2018. WHO global status report on road safety 2018. World Heal.
Accidents in India 2018, Ministry of Road Transport and Highways ,Transport Organization, Geneva, pp. 1–424.
Research Wing, Government of India.