CAR-TO-PEDESTRIAN CONTACT SITUATIONS IN NEAR-MISS INCIDENTS AND REAL-WORLD
ACCIDENTS IN JAPAN
Yasuhiro Matsui
Kunio Takahashi
Ryoko Imaizumi
Kenichi Ando
National Traffic Safety and Environment Laboratory
Japan
Paper Number 11-0164
ABSTRACT cars at intersections or on straight roads. We
calculated the TTC from the velocity of a car with an
The number of traffic deaths in Japan was 4,863 in installed drive recorder and the distance between a car
2010. When looking at the number of the road and a pedestrian at the moment a pedestrian initially
accident fatalities (4,863) in 2010, it reveals that appeared on a motion picture captured by the drive
pedestrians account for the highest number (1,714, recorder. As a result, the average TTC was 1.7
35%). To reduce the severity of injuries and the seconds (SD 1.3 seconds). The average TTC was 1.8
number of deaths, active safety devices providing seconds in cases that pedestrians were walking across
pedestrian detection are considered to be ones of the a crosswalk, which was longer than the average TTC
effective countermeasures. The detailed features of the 1.4 seconds in the cases that pedestrians were walking
contact scenarios in car-to-pedestrian are necessary to across the roads without a crosswalk. The authors
develop the safety devices. Since the information on propose that the specifications of the safety device for
the real-world accidents was limited, the authors the pedestrian detection and for automatic braking
focused on the near-miss scenarios captured by drive should reflect the detailed information including the
recorders installed in passenger cars. TTC obtained by the near miss situations, in which the
worst situation was assumed that the cars were
The first purpose of the present study is to ascertain moving toward pedestrians without braking due to car
the utility of using near-miss scenarios for driver's inattentiveness.
understanding the features of the contact situations
between cars and pedestrians. In the present study, the INTRODUCTION
authors investigated the similarities between the data
of near-miss incidents including motion pictures The number of traffic deaths in Japan was 4,863 in
captured by drive recorders and the data of national 2010. When looking at the number of the road
traffic accidents based on real-world fatal pedestrian accident fatalities (4,863) in 2010, it reveals that
accidents in Japan. This study used 163 motion pedestrians account for the highest number (1,714,
pictures of near-miss car-to-pedestrian incident data 35%)1). The Japanese government has an aim to
collected by the Society of Automotive Engineers of reduce the annual fatality count to less than 2,500 till
Japan (J-SAE) from 2005 to 2009. The results 20182)．For example, since head injuries were the most
indicated that 70% pedestrians at intersections or on common causes of pedestrian deaths in
straight roads were crossing the roads in front of the car-to-pedestrian accidents, the Japanese Ministry of
forward moving cars both in accidents and near-miss Land, Infrastructure, Transport and Tourism (J-MLIT)
incidents. Considering the features of pedestrians’ started to assess the safety performance of the bonnet
behaviors from this result, the authors found top of a vehicle. To reduce the severity of injuries and
similarities between accidents and near-miss incidents. the number of deaths, active safety devices such as the
It was made clear that one could estimate the crash severity mitigation system using sensors for
situations in pedestrians’ accident from the near-miss pedestrian detection is regarded as an effective
incident data which included motion pictures countermeasure. Currently, cars with the crash
capturing pedestrian behaviors. severity mitigation system with a sensor of a stereo
camera and automatic braking installed are produced
The second purpose of the present study is to estimate in Japan3, 4). Those cars are expected to be developed
the time to collision (TTC) from the near-miss in consideration of car-to-pedestrian contact situations
incident data. This study analyzed 103 near-miss including the time to collision (hereafter referred to
car-to-pedestrian incident data in which pedestrians TTC). However, the contact situations in accidents
were crossing the roads in front of the forward moving have not been made clear so far, because the detailed
Matsui 1
information on the real-world accidents is limited. For NEAR-MISS IN-DEPTH DATA
example, Rosen et al.5) investigated the pedestrian
locations and car locations one second prior to their J-SAE has collected near-miss incidents data6)
impacts that resulted in fatal accidents, but there have consisting of forward movie pictures, and the cars’
been few other representative examples. Hence this velocities, accelerations and braking signals which are
study focused on the near-miss incidents captured by obtained from drive recorders which has been
drive recorders installed in passenger cars. installed in over one hundred taxis in Tokyo from
2005. The drive recorder was installed on the inner
The near-miss incident is the situation that a car side of the front glass and consisted of a camera and
accident involving a pedestrian is avoided by the three dimensional accelerometers. The near-miss data
attention and braking of a driver. Near-miss incidents include contact events of car-to-car, car-to-pedestrian,
occurred more frequently than accidents. Recently, car-to-bicycle, and car-to-motorcycle impacts.
drive recorders were installed in taxis in Tokyo area
for the purpose of investigating causes of car Table 1 Vehicle-to-pedestrian near-miss data
accidents and educating car drivers. The data of the Daytime
drive recorder consist of forward moving pictures (n=77 incidents)
captured by its camera, and the cars’ velocities, Adult or Pedestrian moving
Location Crosswalk Number
accelerations, and braking signals. If the near-miss child direction
incidents were similar in the feature to the accidents, Straight
None Go straight 4
we determined that the car-to-pedestrian contact （n=24） Across 20
road
Yes
situations or the TTC could be estimated from the （n=28） Across 4
（n=4）
near-miss incidents. So the authors analyzed the None Go straight 1
Adult
near-miss incident data captured by drive recorders Intersection （n=4） Across 3
（n=72）
installed in taxis. （n=42） Yes Go straight 2
（n=38） Across 36
The first purpose of the present study is to ascertain Other* None
Across 2
the utility of using near-miss situations for （n=2） （n=2）
understanding the features of contact situations Child
Straight
None
between cars and pedestrians. In this study, the （n=5）
road
（n=5）
Across 5
（n=5）
similarities were investigated between the data of Other*: in parking areas
near-miss incidents including motion pictures
captured by drive recorders and the data of national
Nighttime
traffic accidents based on real-world fatal pedestrian (n=86 incidents)
accidents in Japan.
Adult or Pedestrian moving
Location Crosswalk Number
child direction
The second purpose of the present study is to estimate None Go straight -
Straight
the TTC from the near-miss incident data so as to help （n=29） Across 29
road
develop the crash severity mitigation system of active （n=32） Yes
Across 3
safety cars in the future. This study analyzed Adult （n=3）
（n=83） None
near-miss car-to-pedestrian incident data in which Across 3
Intersection （n=3）
pedestrians were crossing the roads in front of the （n=51） Yes Go straight -
forward moving cars at intersections or on straight （n=48） Across 48
roads. The authors calculated the TTC from the Straight
None
velocity of a car with an installed drive recorder and road
（n=1）
Across 1
Child （n=1）
the distance between a car and a pedestrian at the （n=3） Intersection None
moment a pedestrian appeared on a motion picture Across 2
（n=2） （n=2）
captured by the drive recorder. The worst situation
was assumed when a car was moving toward a
pedestrian without the attention and braking of the car
driver. In the present study, the authors used and The drive recorder’s collection of data is triggered by
analyzed the near-miss car-to-pedestrian incident data a driver’s sudden braking of over 0.5 G deceleration,
by the Society of Automotive Engineers of Japan and the recorder can keep capturing the data for 10
(J-SAE) from 2005 to 2009 in order to estimate the seconds beforehand and 15 seconds after the
TTC considering the worst situation. triggering. In the present study, the authors used 163
near-miss car-to-pedestrian incident data from 2005 to
2009 consisting of 77 incidents in daytime and 86
Matsui 2
incidents in nighttime. Table 2 summarizes the A B Cross in front of the moving car
age-group (adult or child), near-miss incident location C D Go to the same direction as the moving car
(straight road or intersection), crosswalk, and
pedestrian moving direction. In reviewing Table 2, it B B
is seen that the majority of the pedestrians were
C C
crossing the roads in front of forward moving cars at D D
intersections or on straight roads regardless of whether
A A
it was daytime or nighttime.
C D D
C
CONTACT SCENARIOS IN REAL-WORLD B B
ACCIDENTS AND NEAR-MISS INCIDENTS
To clarify the utility of using near-miss
car-to-pedestrian incident data, the authors
investigated the motion pictures captured by drive (1) on a straight road (2) at an intersection
recorders of 163 near-miss incidents from 2005 to Fig. 1 Relationship of moving directions between a
2009 and the national traffic accidents records based vehicle and pedestrian
on 12,283 real-world fatal pedestrian accidents from
1999 to 2003 in Japan7). The relationship of moving Straight
directions between vehicles and pedestrians on road Intersection
straight roads and intersections is defined as shown in A B Cross in front of the moving car
Figure 1. On the straight roads, the cases that
C D Go to the same direction as the moving car
pedestrians were crossing the roads in front of the
forward moving cars were defined as “A”, and the
100
cases that pedestrians were walking at the same D 18% 21%
19% 31%
direction with the moving cars were defined as “C”. 80
C 16% 5%
At the intersections, cars moved in three directions:
60 32%
forward, turning right, and turning left. The cases that 35% 30%
B 29%
pedestrians were crossing the roads in front of the 40
forward moving cars were defined as “B”, and the
cases that pedestrians were walking at the same 20 46%
A 38% 39% 38%
direction with the initial moving cars which later
0
turned right or left were defined as “D”. Fatal Near-miss Fatal Near-miss
accident incident accident incident
The distribution of moving directions between
vehicles and pedestrians in accident data and Daytime Nighttime
near-miss data is shown in Figure 2. When focusing Fig. 2 Distribution of moving direction between a
on the distribution ratio of the cases pedestrians were vehicle and pedestrian in accident data and near-miss
crossing the roads in front of the forward moving cars data
(“A” and ”B” in Figure 1), it is seen that these were
67% (fatal) vs. 74% (near-miss) in daytime, and 78%
(fatal) vs. 69% (near-miss) in nighttime, respectively. NEAR-MISS IN-DEPTH ANALYSIS
The results indicated that approximately 70%
pedestrians at the intersections or on straight roads Near-miss in-depth data
were crossing the roads in front of the forward moving In this section, the time to collision (TTC) is estimated
cars. Considering the features of pedestrians’ from the near-miss car-to-pedestrian incident data in
behaviors from this result, similarities are observed which pedestrians were crossing the roads in front of
between accidents and near-miss incidents. It was the forward moving cars at the intersections or on
determined that onecould predict the pedestrian straight roads. Basically, the near-miss incident was in
accident situations by analyzing the near-miss incident such a situation that the car accident was avoided due
data containing motion pictures capturing the to the attention and braking of the car driver. In the
pedestrian behaviors. Therefore, in the next section, present study, the TTC was estimated from the
the authors investigate the detailed situations from the near-miss data considering the worst case that a car
near-miss incident data that cars and pedestrians moving toward a pedestrian would result in accident
approached each other. without the car driver’s braking.
Matsui 3
Table 2 Summary of in-depth analysis of vehicle-to-pedestrian near-miss data (n=103 incidents)
Location Straight road (n=55 incidents) Intersection (n=48 incidents)
Daytime Nighttime Daytime Nighttime
Time Total
(n=28) (n=27) (n=23) (n=25)
Across from Right Left Right Left Right Left Right Left
Adult pedestrian 10 14 12 14 12 11 13 10 96
Child pedestrian 2 2 0 1 0 0 0 2 7
Total 12 16 12 15 12 11 13 12 103
The near-miss incident data that cars and pedestrians LL
approached each other were selected for the purpose
of analysis. As a result, 103 out of 163 near-miss
car-to-pedestrian incident data were used; pedestrians
were crossing the roads in front of the forward moving
cars at the intersections or on straight roads. The age x
group and near-miss location of the 103 incidents are
L
summarized in Table 2. The numbers of the near-miss
incidents were similar in daytime (28 cases) and night 0.85m y
time (27 cases) at straight road and in daytime (23
cases) and night time (25 cases) at intersections. The Forward distance = L
numbers of incidents involving adult and child Ld Lateral distance (Ld) = LL - 0.85
pedestrians were 96 and 7 cases, respectively.
Calculation of TTC Fig. 3 Definition of a lane and distance
The TTC (second) was calculated by the following
formula using the velocity (V: m/s) of a car with an
installed drive recorder and the forward distance (L: Figure 4 shows the estimated TTC distribution of the
m) between a car and a pedestrian at the moment a lateral distance (Ld) from the one side of a car to the
pedestrian appeared on a motion picture captured by pedestrian at the moment that the pedestrian appears
the drive recorder as shown in Figure 3. on a motion picture captured by the drive recorder.
TTC = L/V [1] The TTCs ranged from 0.5 seconds to 5.0 seconds. In
Here, V is the running velocity of the car just before determining the location of a pedestrian relative to the
the driver applies the brake after realizing the center of a car, it is observed that 49 cases were on the
existence of a pedestrian. It was determined whether a right hand side and 54 cases on the left hand side. The
driver applied the brakes by checking the braking average TTC was 1.8 sec (SD 1.5 sec) for the cases on
signal and deceleration signal recorded in the drive the right hand side, and 1.6 sec (SD 1.0 sec) for the
recorder. cases on the left hand side. Since the average TTC
was similar on both sides, the following analyses were
The authors also investigated the lateral distance (Ld: performed regardless of whether the pedestrian was
m) between one side of the car and the pedestrian by located on the right or left hand sides. The average
using the following formula. TTC was 1.7 sec (SD 1.3 sec) for the total 103 cases.
Ld = LL - 0.85 [2]
Here, LL (m) is approximately 1.7m which is the The distribution of the estimated TTC and forward
distance between the center of the drive recorder distance (L) between a car and a pedestrian is shown
camera (the center of the car) and the pedestrian, and in Figure 5. Looking at the figure, one could observe a
0.85 m of the half distance of the full width of the car. linear correlation between the forward distance and
TTC theoretically.
Matsui 4
Left Right The distribution of the estimated TTC and the car
5.0 n=54 n=49 running velocity (V) is shown in Figure 6.
Theoretically, the TTC became shorter if the car
4.0 running velocity was getting higher. On the other hand,
no linear correlation between the car running velocity
TTC (sec)
3.0 and TTC is observed. One could speculate several
reasons for the widely-scattered coordinates in Figure
2.0 6. Therefore, in the next section, the authors
investigate the detailed features of pedestrian
1.0 behaviors.
Detailed Feature of Pedestrian Behaviors
10 5 0 5 10 The authors classified the pedestrians’ popping out
Ld: Lateral patterns in front of the drive recorder installed in a car
distance (m) into four categories as shown in Table 3. The
classified patterns are (1) unobstructed view, (2) from
behind a building, (3) from behind a parked vehicle,
and (4) from behind a moving vehicle. The average of
Fig. 4 Distribution of the estimated TTC and lateral
the TTC, the forward distance between a car and a
distance (Ld)
pedestrian, and the car running velocity in classified 4
pedestrian pop out patterns are shown in Figure 7.
Looking at the average TTC, the unobstructed view
5 (1) was longest as 2.9 seconds, which was presumably
caused by the longer forward distance (Ave 16.2m)
4 regardless of the relatively higher running velocity of
a car (Ave 30.3 km/h). The average TTC 1.4 seconds
TTC (sec)
3 from behind a building (2) was similar to the average
TTC 1.3 seconds of from behind a parked vehicle (3).
2 And from behind a moving vehicle (4) was the
shortest at 1.2 seconds, which was presumed to be
1 caused by the higher running velocity of a car (Ave
32.9 km/h).
0
0 10 20 30 40 50 The average of the TTC, the forward distance between
a car and a pedestrian, and the car running velocity in
L: Forward distance (m) classified two pedestrian locations at crosswalk or
Fig. 5 Distribution of the estimated TTC and the without crosswalk of each classification is shown in
forward distance (L) Figure 8. The average TTC in the cases that
pedestrians were walking across a crosswalk was
longer than the average TTC in the cases that
5 pedestrians were walking on the roads without a
crosswalk. It was assumed to result in the effect of a
4 crosswalk that a car driver would have enough time to
put on the brake for the crosswalk ahead. When one
TTC (sec)
3 focuses on the location of a crosswalk, it is observed
that the average TTC 1.9 seconds in cases at
2 intersections was longer than the average TTC 1.6
seconds in cases on straight roads. It was assumed that
1 a car driver would recognize that he or she were in
more danger of hitting a pedestrian at an intersection
0 than on a straight road. So, it was determined that a
car driver would put on the brake and have enough
0 10 20 30 40 50 time to reach a crosswalk at an intersection. On the
V: Car running velocity (km/h) other hand, the average TTC in cases that pedestrians
Fig. 6 Distribution of the estimated TTC and the car were walking across the roads at an intersection
running velocity (V) without a crosswalk was shortest as 1.2 seconds in the
Matsui 5
four categories described in Figure 8. Even though DISCUSSION AND CONCLUSIONS
these were near-miss incidents, it revealed that this In the present study, the authors investigated the
condition had the high potential of causing an accident utility of near-miss situations for understanding the
at an intersection without a crosswalk. features of the contact situations between cars and
pedestrians, and estimated the time to collision (TTC)
Table 3 Classified 4 pedestrian popping out patterns in by focusing on the near-miss incident data. Basically,
front of a car installing the drive recorder the near-miss incident was in such a situation that a
(1) (2) (3) (4) car accident was avoided by the attention and braking
Unobstructed view From behind
a building
From behind
a parked vehicle
From behind
a moving vehicle
of the car driver.
1) The similarities between the data of near-miss
incidents including motion pictures captured by drive
recorders and the data of national traffic accidents
based on real-world fatal pedestrian accidents in Japan
were investigated. The results indicated that 70%
pedestrians at intersections or on straight roads were
crossing the roads in front of the forward moving cars
both in accidents and near-miss incidents. Considering
Car
installing the features of pedestrians’ behaviors from this result,
drive
recorder similarities were found between accidents and
near-miss incidents. It was determined that one could
estimate the situations in car-to-pedestrian accidents
Average TTC from the near-miss incident data which included
Average velocity of a car motion pictures capturing pedestrian behaviors.
Average forward distance of a car & a pedestrian 2) The authors analyzed 103 near-miss
& a pedestrian (m)
2.5 32.9km/h 35 car-to-pedestrian incident data in which pedestrians
Ave forward distance of a car
30.3km/h
Ave Velocity of a car (km/h)
2.0 25.8km/h 30 were crossing the roads in front of the forward moving
24.3km/h
Ave TTC (sec)
2.0sec 25 cars at intersections or on straight roads. In the present
1.5
16.2m
20 study, it should be noted here that the TTC could be
1.4sec 1.3sec
1.0 1.2sec 15 estimated from the near-miss data considering the
8.6m 8.8m
11.1m 10 worst case that a car moving toward a pedestrian
0.5 5 would result in accident without the car driver’s
0 (1)
0 braking. The TTC was calculated from the velocity of
(2) (3) (4)
Unobstructed From behind From behind From behind a car installing a drive recorder and the distance
view a building a parked vehicle a moving vehicle between a car and a pedestrian at the moment a
(n=55) (n=8) (n=28) (n=10)
pedestrian initially appeared on a motion picture
Fig. 7 Averages of TTC, forward distance between a captured by the drive recorder. From the results, the
car and a pedestrian, and the car running velocity in average TTC was 1.7 seconds (SD 1.3 seconds).
classified 4 pedestrian popping out patterns When one focuses on the pedestrians popping out
patterns in front of the cars, it is observed that the
Average TTC
average TTC was the shortest at 1.2 seconds in the
Average velocity of a car cases that pedestrians popped out from behind moving
Average forward distance of a car & a pedestrian
vehicles on the opposite lane. The authors propose
& a pedestrian (m)
2.5 33.0km/h 35 that the specifications of the safety device for the
Ave forward distance of a car
30.0km/h
Ave Velocity of a car (km/h)
2.0
27.7km/h 30 pedestrian detection and automatic braking should
Ave TTC (sec)
1.9sec 20.9km/h 25 reflect the detailed information including the TTC
1.5 1.6sec 20 obtained by the near miss situations.
1.4sec
1.0 15.6m
1.2sec 15 3) In the present study, the authors focused on the
14.1m
10.8m
10 103 near-miss car-to-pedestrian incident data in order
0.5 7.8m 5 to obtain the TTC. Since the feature of the 103
0 0 near-miss car-to-pedestrian incident data was similar
N=7 N=43 N=47 N=4
Straight road Intersection Straight road Intersection to the feature in accident records, the authors could
Crosswalk w/o crosswalk
define the available near-miss incident level for
estimating accident situations based on the present
Fig. 8 Averages of the TTC, distance of a car and a
analysis results such as the average TTC was 1.7
pedestrian, and velocity of the drive recorder installed
seconds (SD 1.3 seconds).
in a car on roads with or without crosswalk
Matsui 6
ACKNOWLEDGMENTS [4] S Kuzumaki, Our Approach to a Safe Sustainable
The authors appreciate Mr. Katsumi Moro, formerly at Society, Journal of Society of Automotive Engineers
the Society of Automotive Engineers of Japan (J-SAE), of Japan Vol.63, No.12, pp.11-19, 2009, paper number
currently with the Tokyo University of Agriculture 20094737.
and Technology, for his cooperation with the analysis [5] E. Rosen, J. Kalhammer, D. Eriksson, M.
using near-miss data. Nentwich, R. Fredriksson and K. Smith, Pedestrin
Injury Mitigation by Autonomous Braking, 21st
REFERENCES International Technical Conference on the Enhanced
Safety of Vehicles (ESV) 2009, paper number
[1] Institute for Traffic Accident Research and Data 09-0132.
Analysis of Japan (ITARDA), ‘Annual Traffic [6] M Nagai, Current Situation in Registration and
Accident Report in 2010’, Tokyo, 2011 (in Japanese). Application of Drive Recorder Data Base, Forum in
[2] Cabinet Office, White Paper on Traffic Safety, JSAE Spring Convention in 2009 “What the Drive
2009. Recorder can contribute for Traffic Safety”, pp.31-40
[3] E Shibata, Development of Driving Assist System 2009.
“EyeSight” by New Stereo Camera, Journal of Society [7] Institute for Traffic Accident Research and Data
of Automotive Engineers of Japan Vol.63, No.2, Analysis of Japan, ITARDA Information No.53 (in
pp.93-98, 2009, paper number 20094102. Japanese), Tokyo, 2004.
Matsui 7
