Hindawi

Journal of Advanced Transportation

Volume 2022, Article ID 1188089, 13 pages
https://doi.org/10.1155/2022/1188089

Review Article
A Systematic Review of Autonomous Emergency Braking System:
Impact Factor, Technology, and Performance Evaluation

Lan Yang ,1 Yipeng Yang ,1 Guoyuan Wu ,2 Xiangmo Zhao ,1 Shan Fang ,1

Xishun Liao ,2 Runmin Wang ,1 and Mengxiao Zhang 1
1
School of Information and Engineering, Chang’an University, Xi’an, China
2
Center for Environmental Research and Technology, University of California, Riverside, CA, USA

Correspondence should be addressed to Yipeng Yang; yyp1048126411@163.com and Guoyuan Wu; gywu@cert.ucr.edu

Received 13 September 2021; Revised 27 February 2022; Accepted 11 March 2022; Published 18 April 2022

Academic Editor: Francesco Galante

Copyright © 2022 Lan Yang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
In order to track the research progress of AEB-related technologies, this paper makes a systematic analysis and research on the
impact factors, key technologies, and effect evaluation of AEB. First, the paper deeply analyzes the three levels of factors affecting
the performance of AEB, which are vehicle factors, driver factors, and environmental factors. Second, the paper deeply studies the
technical status of the three subsystems of environment perception, decision-making, and control execution. Particularly, the
performance of Mazda, Honda, NHTSA, Berkeley, and Seungwuk Moon are compared and analyzed based on MATLAB. Third,
the paper summarizes the current AEB virtual test methods, closed field test methods, and its test sites. Three classic evaluation
methods in the world, including the AEB test evaluation standards of ENCAP, IIHS, and i-Vista are analyzed. Finally, the paper
prospects the specific research directions, including the protection of vulnerable road users, target detection method, collision
avoidance strategy, complex scenarios application, and application of emerging technologies.

1. Introduction pedestrians, or other traffic participants at the front. Then,

the system automatically triggers the actuator to implement
The continuous advancement of the global economy has necessary braking to avoid the collision or mitigate its se-
increased the number of vehicles every year. The World verity. The working process of the AEB system can be di-
Health Organization (WHO) report shows that more than vided into the following three stages. (1) Normal stage: the
50 million people are injured and approximately 2.5% of vehicle will not collide with the vehicle at the front or other
them die in road traffic accidents every year [1]. According to obstacles (pedestrians, cyclists, etc.); the AEB system will not
the statistics, approximately 90% of the total road traffic intervene in the driving behavior. (2) Early warning stage:
accidents are caused by drivers’ operating errors due to the AEB system will alert the driver immediately through
inexperience and irregular driving behaviors (drunk or fa- visual or audial warning sign, or by tightening the safety belt.
tigue driving). In most of the cases, the driver is unaware of (3) Braking stage: the AEB system uses a single-stage or
the risk of collision, or the reaction time is too short to deal multistage braking strategy (i.e., directly applies the maxi-
with the imminent collision properly. mum braking pressure or gradually increases the braking
As the sensor and control technologies progress con- pressure) to avoid the collision.
tinuously, the ADAS allows drivers to identify potential With the advancement in technology, the application
dangers in different scenarios promptly, thereby improving scope of AEB systems remains expanding and their effec-
driving safety. As a prominent example, the AEB system tiveness of collision avoidance keeps improving. In recent
leverages on-board sensors (such as millimeter-wave radar years, researchers worldwide have proposed various AEB
and/or camera) to perceive the downstream traffic condition systems for pedestrians [2], cycles [3, 4], motorcycles [5–9],
and evaluate the potential collision risk with remote vehicles, electric vehicles [10, 11], large buses [12], special vehicles
2 Journal of Advanced Transportation

[13, 14], and other target and application objects and have The impact of system factors on the collision avoidance
applied emerging technologies such as machine learning effect is reflected in the system error [26], braking delay [27],
[15–17] and vehicle-to-vehicle (V2V) communications [18] maximum deceleration, and control strategy [28]. System
to the latest AEB systems to improve their accuracy. The errors are caused by measurement errors and incomplete
AEB systems can reduce the number and/or severity of environment perception, which leads to uncertainty in
relevant crashes, resulting in reduced number of traffic fa- system decision-making. Braking delay refers to the time
talities. Several existing studies have shown that AEB could between the driver pressing the brake pedal and the brake
reduce rear-end collisions by 25% to 50% [19–22]. system starting to build up pressure, which depends on the
In this study, extensive literature on AEB systems is ability of the brake execution system. When the vehicle
collected and summarized from three aspects: its impact adopts full-braking behavior, greater vehicle deceleration
factors, system structure (consisting of environment per- will shorten the braking time and ensure the safety of the
ception subsystem, decision-making subsystem, and exe- vehicle, but too high deceleration will affect the driver’s
cution subsystem), and its test and evaluation (as shown in driving experience. With the advancement in sensor and
Figure 1). The remainder of this paper is organized as fol- braking technologies, the AEB system’s ability to recognize
lows. Section 2 analyzes the various factors that affect the targets, the amount and accuracy of data obtained, and the
system performance. The system structure was introduced, system delay time are continuously improving. The choice of
and some collision avoidance algorithms were simulated and control strategy directly affects the overall performance of
compared in Section 3. Section 4 describes different AEB test the AEB system, which is the core part of the system.
and evaluation methods employed in different countries, Comparing the current widespread collision avoidance al-
and Section 5 concludes this paper and presents future gorithms based on safe distance or time-to-collision (TTC),
research direction. when the vehicle speed is lower than 60 km/h, the collision
avoidance algorithm based on TTC has a better stability and
2. Impact Factors smaller braking force, providing the driver a more com-
fortable driving experience. When the vehicle speed exceeds
The AEB system performance is affected by both the intrinsic 60 km/h, the collision avoidance algorithm based on safe
and extrinsic factors of the equipped vehicle while driving. The distance can ensure the reliability of vehicle emergency
intrinsic factors cover on-board sensing, decision-making as braking [29].
well as actuation, while the extrinsic factors deal with other
than the vehicle itself. This study further divides these factors 2.2. Driver Factor. The “autonomous” of the AEB system
into three categories according to the level of influence: vehicle, primarily means that the driver’s intervention is not re-
driver, and environment factors. These factors may coexist quired. However, each driver’s driving style varies owing to
during the AEB working process to affect safety, comfort, and his or her characteristics, such as driver’s age, gender, ex-
energy consumption as shown in Figure 2. perience, responsiveness, and psychological endurance.
Therefore, unified collision avoidance logic and evaluation
criteria may not be applicable to different drivers. Therefore,
2.1. Vehicle Factor. While driving, vehicles may encounter
in addition to ensuring safety, the driver’s driving comfort
complex situations in terms of road environment and traffic
should be guaranteed to a certain extent.
conditions, such as (horizontally and vertically) curved
To analyze the driving styles of different types of drivers
roads, intersections, overtaking, and lane changes. At
and avoid the dissatisfaction and doubt caused by the control
present, AEB system is generally applicable to low- and
strategies that do not conform to their driving styles, drivers
medium-speed traffic scenes, and high-speed scenes are
are classified according to the true test data of driving
rarely involved. In the international test standard, the
characteristics [30]. This data can be obtained by a driving
maximum speed of the test vehicle is 80 km/h. Meanwhile, it
characteristic identification model based on a hidden
is important to identify the most dangerous target in the
Markov chain [31] or a Bayesian filter and support vector
shortest time under complex road conditions. The basic
machine model [32]. Drivers can be divided into three types:
function of the AEB is that the object in front can be rec-
radical, standard, and conservative [33]. After determining
ognized by the camera and radar. However, as the main
the driver style, different control strategies and parameters
sensing equipment of vehicle, cameras and radars are less
can be set for different types of drivers to enhance the control
effective in bad weather and poor light conditions, such as
accuracy and driving comfort of the system [34].
sandstorms, fog, snow, and darkness. The field-of-view
(FoV) angle of the sensor has a significant effect on avoiding
collisions, particularly for collisions between vehicles and 2.3. Environment Factor. The environment factors that affect
pedestrians (or bicyclists). Studies have shown that when the the system performance are external factors, including
detection angle of the AEB system is set to 30° to 50°, more weather, light, and road conditions, in addition to the vehicle
than 95.3% of severe injury and death accidents and 78.5% to itself and driver. Roads can be majorly divided into ordinary
92.2% of minor injury accidents can be detected [23]. With road, cross road, and tee sections road. The road conditions,
an increase in the detection angle, more targets (specifically including the road adhesion coefficient, slope, and type,
pedestrians and bicyclists) can be detected [24], avoiding directly influence the efficiency in achieving the expected
more accidents [25]. effect after implementing the braking action.
Journal of Advanced Transportation 3

Impact Factor
Environmental
Vehicle Factor Driver Factor
Factor

AEB System
Environment Execution
Perception
Decision-making
Sensor Early
Information Warning
Vehicle

Input Output
Environment Information Braking
Target Risk Decision Command
Information Pedestrian
Recognition Assessment Control

Ego Vehicle Steering

Bicycle
Information

Test and Evaluation

Figure 1: AEB system framework.

Impact Factors Vehicle Speed Performance

Field of View

Sensor Type

Vehicle System Error Safety

Braking Delay

Control Strategy

Maximum Deceleration
Impact Factors
Driver Comfort
of AEB
Driving Style

Driver Age

Driver Gender

Energy
Environment Weather & Light
Consumption

Road Type

Road Adhesion
Coefficient

Road Slope

Figure 2: Impact factors of AEB performance.

Weather and light are easy to distinguish for drivers, system perception mechanism to the surrounding objects
while AEB systems mainly depend on sensors and are and the deceleration effect of the vehicle after braking will be
susceptible to weather and light. Under special conditions affected. Weather can be divided into sunny, cloudy, rainy,
(e.g., rainy, snowy, and fog), the perception ability of the or severe weather. Light can be divided into daytime and
4 Journal of Advanced Transportation

night with and without street lights. The proportions of good driver factor is the focus of current research. On the premise
light conditions among fatalities and injuries were calculated of ensuring the safety of AEB system, improving the driving
to be 75.58% and 85.51%, respectively, except for collisions experience of different drivers can further improve the
that occur on roads without streetlights. AEB system is performance and reliability of AEB system.
assumed to only work effectively in collisions that occur in
good weather conditions, including sunny and cloudy days. 3. Subsystems of AEB System
The proportions of good weather conditions among fatalities
and injuries were calculated to be 88.36% and 88.82%, The pipeline of the AEB system primarily includes three
respectively. components: environment perception subsystem, decision-
The road adhesion coefficient is affected by the road type, making subsystem, and execution subsystem. The envi-
wetness, tire property, and air pressure. This can be con- ronment perception subsystem is to obtain the vehicle and
sidered as the static friction coefficient between the tire and surrounding road information through on-board sensors,
pavement. If the AEB system does not consider the effect of such as cameras, radars, and thermal sensors, and send the
pavement adhesion for a vehicle travelling on a road with a information to the decision-making subsystem. Based on the
small adhesion factor, the braking distance will be larger received information, the decision-making subsystem judges
than expected, and the collision avoidance efficiency of the the critical situation of the current road conditions and
system will be reduced. Therefore, to enhance the perfor- simultaneously determines whether early warning, braking,
mance of the AEB system for different pavement adhesion and other collision avoidance strategies (e.g., steering) must
coefficients. Rajamani et al. [35] estimated the peak pave- be implemented. The subsystem then transmits commands
ment friction based on the longitudinal, transverse, and to the execution subsystem for executing the collision
normal direction tire forces. Hwang and Choi [36] and Sevil avoidance operation of the corresponding module.
et al. [37] employed an adaptive AEB system for different
pavement adhesion coefficients. Han et al. [38] and Kogl-
bauer et al. [39] experimentally demonstrated that the 3.1. Environment Perception Subsystem. The environment
adaptive AEB system can alleviate collisions and improve the perception subsystem primarily comprises various sensors
subjective safety and trust of drivers through the driver’s that collect information and identify targets. At present, the
actual driving. most commonly used sensors in AEB systems include
The road slope is mainly divided into uphill and 77 GHz millimeter-wave radar, lidar, mono-binocular
downhill, the AEB system must consider the influence of the cameras, and thermal sensors. These mainstream sensors
vehicle’s gravity effect on braking deceleration. The accel- possess different characteristics. For instance, the millime-
eration component of the vehicle gravity on the uphill ter-wave radar has a better penetration and large detection
section increases the maximum deceleration of the vehicle, range, and is unaffected by light and weather. However, it is
causing the vehicle to complete braking ahead of time. The expensive, and the target recognition is difficult. Response
gravity component of downhill vehicles offsets part of the time of lidar is short with high ranging accuracy, but it is
braking force, reduces the maximum deceleration and in- expensive and can be affected by weather. The cost of the
creases the braking time and distance, along with the monocular camera is low which can effectively identify the
likelihood of collision. Therefore, while considering the target, but the detection range is short.
slope, the dynamic characteristics of the vehicle on the uphill The difficulty of research on environment sensing sub-
and downhill slopes should be analyzed, and the minimum systems of AEB system is the identification and selection of
braking distance and braking time should be calculated by targets, including the accuracy of recognition, radar and
estimating the force applied to the vehicle [40]. To ensure camera data fusion, and the selection of the most dangerous
that the AEB system can effectively avoid collisions under targets. The false positive of the environment perception
different road slope conditions, the existing road slope of the subsystem will cause the false triggering of AEB. Therefore,
vehicle must be determined. To estimate the road slope, Bae in order to reduce the false positive, OEM (original
et al. [41] proposed two methods based on global positioning equipment manufacturer) will generally carry out a large
system; Kim et al. [42] used a combined extended Kalman number of performance tests and field operation test (FOT).
filter based on vehicle sensor information and strong According to previous research [44–47], sensor fusion
nonlinear vehicle dynamics, tire, and inclination angle technology can improve target perception ability in AEB
models; and Vahidi et al. [43] used the recursive least square system. With the research development, the AEB system will
method with a forgetting factor. face more complex traffic environments (curves, intersec-
This section summarizes the vehicle-driver-environment tions, overtaking, lane changes, etc.); therefore, the most
factors that can affect the AEB system. At present, there have dangerous targets in the current and complex road condi-
been a large number of studies on control strategies and road tions must be analyzed. At present, the most dangerous
conditions. However, the factors such as system error, target selection strategy includes three methods: the target
braking delay and maximum deceleration have relatively closest to the vehicle in the current lane, judging the driver’s
little impact on AEB, and fixed values are adopted in most intention to identify the most dangerous target, and selecting
studies. Which leads to that although AEB system has a the most dangerous target according to the driver’s behavior.
mature theoretical basis, these parameters need to be In addition to the use of physical sensors, the rapid
changed when applied to specific models. Considering the development of wireless communication technology (e.g.,
Journal of Advanced Transportation 5

vehicle-to-vehicle communication, V2V) and high-perfor- the evaluation parameters should be set and the widely used
mance computing can improve the AEB’s target detection evaluation parameters are the safe distance and TTC. In
capability. The fundamental benefit of V2V lies in its ability addition, the predicted minimum distance [55], time to
to exchange vehicles’ information, which enables the system brake [56, 57], time to act, time difference to collision [58],
to make better decisions in terms of safety. At present, some and critical speed for decision-making [59] can be used as
studies have proved the effectiveness of V2V technology in evaluation parameters.
vehicle collision avoidance [48], particularly in pedestrian Subsequently, based on the collected information and
target recognition [49]. The problems that limit the devel- predicted trajectory, the parameter threshold of the risk
opment of this technology are the long-term coexistence of scenario is calculated by collision avoidance algorithm.
vehicles with V2V and without V2V on the road. Owing to the algorithm is directly related to the execution
time of braking, it has a great impact on the collision
avoidance ability of the AEB system. In section 3.2.4, several
3.2. Decision-Making Subsystem. Decision-making subsys- common collision avoidance algorithms are simulated and
tem, the core part of the AEB system, is also the focus of most compared, which take the safety distance as the evaluation
existing research. It includes three main parts: target rec- parameter. Finally, a risk assessment is performed and de-
ognition, risk assessment, and decision-making control fined quantitatively. Risk quantification is an intuitive ex-
strategy. pression of the degree of risk. Shimizu et al. [60] developed a
risk quantification model based on collision speed in a
3.2.1. Target Recognition Strategy. According to different dangerous scene. The effectiveness of the model in pre-
collision avoidance targets, the AEB system recognizes dicting collision risk was proved by comparing the simu-
objects into three categories: vehicles, pedestrians, and bi- lation results with the actual driving data. Cafiso et al. [61]
cyclists. Pedestrian and bicyclist are mainly focused for used the pedestrian risk index to assess the potential severity
target recognition. Owing to the different conditions of of danger and duration. For instance, the risk coefficient ε is
pedestrian clothing, age, and gender, and difficulty in pre- defined numerically based on the current car spacing d and
dicting the trajectory, the requirement for accurate target safety distance (including warning distance dw and braking
recognition is significantly high. Generally, the speed of distance dbr).
bicyclist is higher than that of pedestrians, and 50% of severe d − dbr
and fatal injuries in bicyclist accidents are observed at the ε� . (1)
dw − dbr
speed range of 12 to 15 km/h [50].
If the pedestrian is considered as a stationary target and When ε > 1 and d > dw , the vehicle is in a safe state, and
there is a possibility of collision, the longitudinal collision the AEB system does not intervene. When 0 < ε < 1 and
avoidance will be considered only. However, to the moving d < dbr, the vehicle may collide, and the driver needs to be
target, the possibility of collision is calculated. Therefore, to reminded to perform collision avoidance operations. Fur-
enhance the detection and tracking ability of AEB system for thermore, when ε < 0 and dbr > d, the vehicle needs to apply
moving targets, ensuring the optimum effect and cost, ul- emergency braking for avoiding a collision.
trasonic array sensor is generally selected [51], which can
achieve 86% detection accuracy, with a detection time of
0.8 s [52]. In addition, Lee et al. [44] proposed a robust 3.2.3. Decision-Making Control Strategy. The decision-
pedestrian tracking method based on a multisensor fusion making control strategy determines the corresponding
strategy and designed the activation area by predicting the collision avoidance operation according to the dangerous
possible distance of pedestrian collision through the braking extent based on existing road conditions. The vehicle is
model, which proved to be advantageous. Song et al. [53] considered as the control target. The existing decision
proposed a new theory and algorithm to predict the position control models include the hierarchical control, brake
of pedestrians and determine accurate warning and braking steering control, acceleration control [62], and brake pres-
times. This algorithm can effectively avoid or mitigate pe- sure adaptive models [63].
destrian collision accidents when the vehicle speed is below The hierarchical control model was designed to deal with
40 km/h. Park et al. [54] proposed a pedestrian target se- complex scenarios such as pedestrian collision accidents. To
lection method based on a funnel graph structure. By estimate the danger more accurately, the upper-level eval-
comparing the predicted position of the pedestrian target uation and control part usually adopt new technologies, such
with the current position, the probability of collision was as fuzzy neural networks. Yang et al. [2] used the hierarchical
estimated, and the effectiveness of the method was verified control model for upper fuzzy neural network and lower
through simulation and real vehicle tests. proportion-integration-differentiation (PID) control, and
introduced a genetic algorithm to train the fuzzy neural
network. Based on the theoretical basis of TTC and braking
3.2.2. Risk Assessment Strategy. Risk assessment strategy is safety distance, AEB-pedestrian system early warning model
the judgment of the AEB system on the possibility of col- was established, and the credibility of the control strategy
lision and severity of the accident under the current working was proved through experiments. The radial basis function
condition, which provides a reference for the system to neural network (RBFNN)-based variable structure control
perform corresponding collision avoidance operations. First, (VSC) [16] was used to optimize the AEB system to achieve
6 Journal of Advanced Transportation

higher deceleration. Christopoulos et al. [64] proposed a The results show that under the same conditions, the
rear-end collision avoidance control strategy using hierar- braking distance obtained by the Seungwuk Moon algorithm
chical control. The upper risk assessment layer continuously was the most reasonable and the effect was optimum. In
calculated the threat parameters related to braking control to Berkeley algorithm, braking distance distribution was rea-
avoid collisions, and the lower tire slip control layer used sonable and simple, when considering fewer factors. The
RBFNN and VSC. The results showed that the proposed Mazda algorithm was too conservative because its calculated
control scheme can effectively achieve collision avoidance. braking distance is clearly large, which leads to a high alarm
When the safety distance is short, the vehicle may change rate and reduces user’s trust. The Honda algorithm was
its lane to avoid collisions quickly; according to this, several relatively too radical because its calculated braking distance
studies have been conducted on automatic emergency is clearly small, and the system intervention time was
braking systems. Llorca et al. [65] used a stereo vision de- delayed.
tection module to calculate the collision time between the
vehicle and target. A simulation controller was used to
simulate the human behavior and reaction. The collision 3.3. Execution Subsystem. When the AEB decision-making
avoidance system was primarily based on automatic steering. subsystem judges that the vehicle is about to collide, the
The results showed that the system could effectively avoid actuator receives the command to remind and protect the
pedestrian collisions at a speed up to 30 km/h. To solve the driver through warning and braking. In the early warning
problem of over-steering and instability when the vehicle stage, the execution subsystem can form different early
speed is high, Choi et al. [66] proposed a collision avoidance warning strategies through a combination of visual, audi-
strategy of simultaneous steering and braking using the tory, and tactile means. The specific forms are buzzer alarm,
nonlinear model predictive control method, and proposed a human-computer interface image alarm, and tightening or
constraint on the wheel steering angle considering the retracting seat belts. In the braking stage, the current braking
predicted lateral acceleration of the vehicle. pressure building methods can be divided into three types:
the active pressure building of motor piston pump based on
electronic stability control (ESC), active pressure building of
3.2.4. Comparative Analysis of Different Collision Avoidance high-pressure energy storage based on electronic hydraulic
Algorithms. The collision avoidance algorithm is an im- braking, and active pressure building of master cylinder
portant part of the AEB risk assessment strategy. The power-assisted motor based on electronic power-assisted
judgment of the dangerous state and collision possibility braking [76].
is obtained by comparing with the actual data, and the Owing to the disadvantages of high-pressure accumu-
evaluation results directly affect the control behavior of lators, such as sensitivity to vibration, low safety and reli-
the system. At present, several algorithms for estimating ability, large volume, and high cost, auto parts
the severity of vehicle rear-end accident safety assessment manufacturers select motor piston pump, master cylinder
models exist. The specific evaluation parameters can be power motor, or a combination of motor piston pump and
divided into safety distance, TTC [67], time headway [68], master cylinder power motor as the active pressure building
minimum deceleration [69], and driver perception models scheme. At present, the active pressure building of motor
[70]. Currently, there are several applications of AEB piston pump is widely used as an actuator in AEB systems.
products based on the safety distance and TTC models. However, the hardware and performance require further
The safety distance model considers the real-time strengthening owing to limited precision and time. There-
headway as the safety evaluation parameter and provides the fore, a brake-by-wire technology can be employed to resolve
judgment basis for further warning and braking by calcu- previously mentioned issues. The brake-by-wire technology
lating the safety critical value of the relative distance between has the advantages of fast braking response, high braking
the ego and front vehicles. The threshold algorithm includes energy, simple structure, and easy expansion. This is a major
five main algorithms: Mazda, Honda, NHTSA, Berkeley, and development direction of the future brake system. To further
Seungwuk Moon (Table 1).This type of algorithm is based on reduce the response time of active braking, shorten the
the speed and acceleration of two vehicles, considering the braking distance and simplify the braking system, the re-
driver reaction time, braking system delay, and other factors. search and development of new generation braking systems,
The calculation method of this critical value is constantly such as electro-mechanical and electromagnetic braking, are
improving. At present, the safety distance model is widely also important directions for the braking scheme selection of
used in the safety assessment of AEB systems. AEB system in the future.
According to the basic threshold calculation method This section summarizes the three subsystems included
of the proposed safety distance model, four algorithms in a complete AEB system. The environmental perception
other than the NHTSA algorithm with complex param- subsystem has the problem of false positives based on the
eters were simulated (as shown in Figure 3). To compare existing vehicle sensors. Reducing false positives and im-
the algorithm characteristics under the same working proving the perception ability are the problems to be solved
conditions, uniform parameters were used: the maximum by AEB system and other vehicle active safety technologies.
acceleration of the ego and target vehicles was 6.5 m/s2, With the development of V2V technology, the vehicle
driver’s reaction and system delay times were 0.5 and 1 s, perception ability will be further improved. The key of the
respectively. decision-making subsystem is the collision avoidance
Journal of Advanced Transportation 7

Table 1: Existing algorithms for the safety distance model.

Research Organization Algorithm Character
Doi [71] Japan Mazda corporation Mazda More scenarios covered
Fujita [72] Honda corporation Honda Two level early warning structure
Burgett et al. [73] NHTSA NHTSA High accuracy of early warning
Seiler et al. [74] University of Berkeley Berkeley Adapting to different road conditions
Moon and Yi [75] Han Yang university Seungwuk Moon Estimation of pavement adhesion coefficient

Mazda algorithm Honda algorithm

50
120
40
Braking Distance (m)

100

Braking Distance (m)

80 30
60
20
40
30 10
20 /s) 30
20 d(
m /s)
0 e 20 (m
10 Spe 0
10 ee d
p
30 25 20 cle 30 25
icl
eS
15 10 5 0 hi 20 15 10 0 h
Relative Speed (m 0
o Ve Relative Speed (m/s)
5 0 Ve
/s) o
Eg Eg
Berkeley algorithm SeungwukMoon algorithm

60 60
50 50
Braking Distance (m)

Braking Distance (m)

40 40
30 30
20 20
30 30
10 /s) 10 /s)
m m
20 d( 20
ed
(
0 10 p ee 0 10 pe
30 25 20 cl eS 30 25 20 cleS
15 10 5 0 hi 15 10 5 0 hi
Relative Speed (m/s)
0
o Ve Relative Speed (m/s)
0
oV
e
Eg Eg
Figure 3: Simulation results of the four algorithms.

algorithm. The experimental results show that different al- 4.1. Test Methods. Currently, the AEB system is a relatively
gorithms have great differences in safety and comfort. The mature active safety system. The specific tests are aimed at
optimization of the algorithm has always been the focus of existing or improved sensors, collision avoidance algorithms,
AEB research. When braking, the execution subsystem actuators, and human-machine interfaces to verify their per-
mostly adopts multilevel braking strategy to improve the formance, stability, robustness, and safety. First, based on real
driver’s comfort. At present, research has added steering road data, a relationship between driving behavior, road, en-
operation on the basis of traditional vehicle braking to avoid vironment, and other traffic participants is analyzed, and the
the collision with short vehicle spacing. test scenario is constructed. Subsequently, the test evaluation
method is designed. At present, two methods are used to test
4. Test and Evaluation AEB systems: virtual and closed field test. Table 3 presents the
comparison results of the two methods.
At present, test and evaluation procedures for AEB systems
are well established all over the world. Since 2013, auto-
mobile safety certification authorities, such as Euro-NCAP, 4.1.1. Virtual Test. Virtual test usually uses various simu-
NHTSA, IIHS, CNCAP, and i-Vista, have included AEB test lation software tools to restore, track, and collect data for
and evaluation procedures (Table 2). specific test scenarios. Vehicle sensors, dynamics model,
8 Journal of Advanced Transportation

Table 2: International standards and test report.

Country/
Organization Standard name Year Document name
Region
Intelligent transport system-forward vehicle collision migration system
ISO International ISO-22839 2013
——operation, performance, and verification requirement
Front crash prevention
IIHS American 2013 Autonomous emergency braking test protocol (version 1)
testing
Objective tests for automatic crash imminent braking (CIB) systems final
NHTSA American — 2014
report
Uniform provisions concerning the approval of motor vehicles with regard
ECE United nations ECE131 2014
to the advanced emergency braking systems (AEBS)
SAE American SAE-J3087 2017 Automatic emergency braking (AEB) system performance testing
i-Vista SM-ADAS-
i-Vista China 2018 Autonomous emergency braking rating protocol
AEBR-A0-2018
European
Euro-NCAP ENCAP-2020 2020 European new car assessment programme (test protocol-AEB system)
union
CNCAP China CNCAP-2021 2021 Test method of active safety ADAS system

Table 3: Comparison of test methods.

Virtual test Closed field test
Scene construction
Controllable scene, low cost, fast speed Controllable scene, high cost, low speed
capability
Scene coverage Define test scene independently, high coverage Restricted by the site facilities, low coverage
Extract holographic test data, comprehensive data, high Extract testing facility data, comprehensive
Data
precision data
Efficiency Support multivehicle collaborative testing, high efficiency Restricted test objects number, low efficiency

controller and driver, as well as traffic environment, are (3) Monte Carlo Simulation: Based on the data collected
simulated by simulation software, and dangerous scenarios from the real vehicle, a random test was conducted
are considered to simulate the AEB system. Currently, through the scene generated by the Monte Carlo
numerous virtual simulation tools are available. The benefits simulation.
of this method include: simplicity, safety, low cost, ease of (4) Accelerated Evaluation [79–81]: Importance sam-
repetition, and enabling of detailed analysis. With these pling theory and the cross-entropy method are used
existing advantages, virtual tests can be adopted in rare and to identify key test scenarios with high precision and
hazardous conditions and/or used to traverse critical values speed, to achieve accelerated simulation and
of key system parameters through repeated tests. However, evaluation.
the reliability of the test results depends on the selection of
test scenarios and accuracy of the simulated vehicles and
environment. 4.1.2. Closed Field Test. Closed field test is used to conduct
The test of the AEB system is functional in nature and real vehicle test on real roads without excessive external
scenario-based. It can be specifically described as a sequence interference, resulting in high controllability, reliability, and
of events occurring in chronological order [77]. The gen- repeatability. However, owing to its high cost, closed field
eration of accident scenarios is mostly based on local ac- testing is widely used in new car assessment program
cident depth investigation data (GIDAS, CIDAS, etc.) and (NCAP) institutions and existing vehicle testbeds in various
real natural driving data. The rapid generation of virtual test countries.
scenarios is the focus of research in the fields of driving In closed field test, it is equally important to select
assistance systems and automatic driving (as shown in representative test scenarios. Appropriate test scenarios can
Table 4). Currently, the methods available are as follows: produce reliable test results and reduce testing costs. At
(1) Test Matrix: The test dimension matrix is con- present, the most commonly used closed field test scenarios
structed based on key parameters (each point of the are those defined in the NCAP regulations of various
matrix corresponds to a scene). The test scenarios are countries where Euro-NCAP provides the basis for the test
sorted and tested accordingly. and evaluation regulations. Based on the actual road traffic
accident databases of the United States, China, and other
(2) Worst-Case Scenario Evaluation [78]: On the basis of
countries, AEB test procedures have been issued for their
monotonic change feature extraction of system
own countries. According to the evaluation procedures and
function, boundary conditions are set and effect
related test scenarios, the AEB test scenarios are divided into
function is optimized to test the most challenging
rear-end collisions, pedestrian accidents, and bicyclist
scenario.
accidents.
Journal of Advanced Transportation 9

Table 4: Partly researches on AEB test scenarios.

Researchers Test scenarios Scenario source Test method
Fleury and Brenac Similar accidents generate typical accident
Rear-end French accident data
[82] scenarios
Pedestrian and vehicle collision at All pedestrians can be detected when FoV is
Huang et al. [83] STRADA
intersections greater than 60°
Niewöhner et al. Propose an improved AEB-pedestrian test
Pedestrian and vehicle collision GIDAS, UDV
[84] method
Lenard and Pedestrians collide with the front and Typical pedestrian accident scenes are obtained
STATS 19, OTS
Danton [85] rear ends of vehicle by comparison
Camp and Lubbe Accident data of five Provide theoretical support for the development
Bicyclist and vehicle collision
[50] European countries of AEB-bicyclist test scenarios
Cluster analysis of AEB test scenarios and
Sander [86] Collision at intersection GIDAS
established evaluation method
Summarized three dangerous scenes of
Chen et al. [87] Pedestrian and vehicle collision CIDAS
pedestrian vehicle collision
Eight typical dangerous scenes are obtained by
Xu et al. [88] Collision at intersection NAIS
hierarchical clustering algorithm

A vehicle comprehensive test site is an indispensable is determined according to the actual reduction in relative
practical condition for closed road tests. Various countries have velocity. In the case of incomplete collision avoidance, the
actively invested in construction and have achieved remarkable linear interpolation method is used to calculate the score of a
results. Table 5 summarizes a list of existing test sites for au- single test velocity. For each test speed above 40 km/h, score
tomobiles that have the ability to test the automobile ADAS will only be awarded when the actual measured test speed is
system. These test sites cover various road types, such as reduced by at least 20 km/h.
highways, urban roads, and rural roads, and a few of them can The score formula for each test speed is as follows:
even emulate rain, fog, and other weather conditions.
Intelligent and connected vehicle has become the de- Vrel_test − Vrel_impact
Scoretest_speed � 􏼠 􏼡 × pointstest_speed , (2)
velopment trend of the industry. As an important link, the Vrel_test
construction of automobile closed test site plays a key role in
accelerating the upgrading of the industry. By learning from where, Vrel_test and Vrel_impact represent the theoretical rel-
the successful experience at home and abroad and allocating ative test speed and actual relative collision speed,
resources reasonably, the automobile test and evaluation respectively.
system can be further improved. The AEB-VRU (vulnerable road user, including pedes-
trian and cyclist) score is determined by the total score of the
three test systems (head, upper limb, and lower limb). The
4.2. AEB System Evaluation Standard. The complete eval- AEB-VRU part can only be scored if the total score is at least
uation method comprises three parts: identification of 18 points. The total score of the pedestrian and bicyclist AEB
evaluation metrics, determination of index weights, and systems is 9 points, respectively. The standard scores of test
selection of evaluation methods. The test data are collected, speed for each test scenario are calculated by multiplying the
processed, and analyzed for evaluating the test process. The score rate by the scenario score.
vehicle systems, modules, or algorithms are then evaluated The score formula for each test scenario is as follows:
using the evaluation method. The key performance index of
the AEB system is the successful rate (i.e., whether the 􏽐 Scoretest_speed
Scorescenario � × pointsscenario , (3)
system can avoid the collision). If the collision cannot be Scoretotal
avoided, then the speed reduction during braking will be
considered as the major evaluation metrics. The evaluation where, Scoretotal represent the total points in each specific
results are quantified and summarized based on the score scenario and pointsscenario represent the score for each
sheet under different test conditions, which represent the scenario.
overall performance of the AEB system in terms of collision
avoidance. Nowadays, different evaluation methods for the
AEB system have been proposed and actively used by major 4.2.2. IIHS Evaluation Standard. Since 2013, the American
institutions, such as ENCAP, IIHS, and CNCAP. Insurance Institute for Highway Safety (IIHS) has included the
evaluation of front crash prevention in the new car evaluation
procedures, and the evaluation level is divided into three levels:
4.2.1. ENCAP Evaluation Standard. The AEB evaluation excellent, advanced, and primary, with a total score of six
standard of ENCAP includes functional tests and human- points. The function of the front crash prevention system is the
machine interface tests. For functional tests, the relative same as that of the AEB system, including the front collision
impact velocity is considered as the key parameter. For test warning and active braking systems. It is mainly evaluated
speed less than 40 km/h, the score weight of each test velocity according to the test vehicle equipment and test results.
10 Journal of Advanced Transportation

Table 5: Partly test sites for vehicles in the world.

Area Road length
Location Name Road type Function
(km2) (km)
M-city 0.129 High/low speed test area 8 Auto-driving and V2X
USA Expressway test area and
Go Mentum Station 8.498 Highway, real hill, viaduct, tunnels, etc. 32.1869
urban test area
Multilane highways, four-street urban areas, V2X, driver behavior, vehicle
Sweden Asta-zero 2 5.7
expressway areas, rural and suburban roads dynamics
UK City circuit 3.04 24 loops 95 Auto-driving, ITS, V2X
Japan Toyota ITS test site 0.035 Intersection — V2V, V2I ICV test
Shanghai Intersection, tunnels, avenues, gas stations and
150 500 Auto-driving, V2X
demonstration zone indoor parking lots
Various special roads, high-speed loop roads The first Road traffic, efficiency,
China i-Vista 2.51
and simulated town roads phase is 6 km information services
Intelligent vehicle test,
CAVTEST 0.28 Urban area road, highway 3.5
C–V2X test

The front crash prevention system has only 1 point there are still some problems in the test of AEB system.
(primary evaluation) when the test is effective and a higher Although closed field test can effectively test the perfor-
evaluation level can be obtained only when the active mance of AEB system, it takes a lot of time and cost and is
braking system is equipped and proved to be effective during not suitable for testing a large number of scenarios. There are
the test. The reduction in vehicle speed during the test few test scenarios specified in international standards, so it is
process is considered as an evaluation parameter. The IIHS impossible to popularize the test to a wider range of road real
stipulates that the test vehicles need to be tested at different conditions. In the future, with the continuous improvement
speeds (19.3 and 40.2 km/h). The score is determined of AEB technology, pedestrian AEB, cyclist AEB test, and
according to the vehicle speed reduction under the action of false response test will be gradually studied and
the system. The more the vehicle speed decreases, the better standardized.
would be the performance of the front crash prevention
system, resulting in lower risk or severity of the collision.
5. Future Work
4.2.3. i-Vista Evaluation Standard. The Intelligent Vehicle A significant amount of effort has been devoted by industry,
Integrated Systems Test Area (i-Vista) has proposed an academia, and government to advance AEB over the past
overall evaluation procedure for intelligent vehicle safety decade. With the breakthrough of hardware technology such
since 2018. The AEB-C2C and AEB-VRU standards are as sensor and brake, the collision avoidance ability of AEB
included in the latest 2020 edition of the regulation. system will be further improved. It is also recommended to
According to the scoring rate, the intelligent safety grade is further explore the following directions in the future:
divided into four evaluation grades: excellent, good, general,
(1) The effectiveness of the AEB function depends on the
and poor.
appropriate selection of potential targets at risk and
The evaluation standard of the AEB-C2C system can be
timely feedback for execution. Therefore, the per-
applied to the forward collision warning (FCW), AEB, and
ception ability of driving environment information is
advanced assistance function tests. The main test scenarios
an important factor restricting the development of
include static, deceleration, and low-speed conditions of the
automatic emergency braking systems. The combi-
target vehicle. Each FCW scenario will be scored if five out of
nation of sensor fusion and vehicle-to-everything
seven tests meet the conditions. Otherwise, no points will be
(V2X) communications can improve the application
scored. In the AEB test scenario where five tests are per-
and sensing range of AEB systems.
formed at the same speed, braking deceleration is used as the
evaluation parameter. Advanced assistance functions in- (2) At present, the AEB is aimed at the front target when
clude the FCW assistance alarm, active safety belt warning the vehicle is running longitudinally, and only few
function, and emergency steering collision avoidance studies exist related to collision avoidance strategy
function. One point for each function is awarded if the under the test scenarios of sudden insertion of a side
function is effective. vehicle, vehicle turning, multivehicle environment,
This section summarizes the test and evaluation methods multipedestrian or riding environment, and limited
and related research of AEB system. At present, various driver vision. Therefore, the AEB collision avoidance
countries and standards organizations are actively pro- strategy in complex multitraffic scenarios should be
moting the popularization of AEB system and constantly studied to optimize its comprehensive performance.
improving the test and evaluation system. Researchers (3) To better cover various test scenarios in a cost-ef-
continue to improve and optimize testing methods and fective manner, augmented reality or mixed reality
speed up the process of testing and evaluation. However, methods should be used to reconstruct, configure,
Journal of Advanced Transportation 11

and enhance the test scenarios. Meanwhile, the Crash Prevention and Injury Control, vol. 14, no. 1, pp. S40–
current evaluation parameters are independent, and S49, 2013.
only the speed reduction is considered as the eval- [7] G. Savino, J. Mackenzie, and T. Allen, “A robust estimation of
uation index. The system can be comprehensively the effects of motorcycle autonomous emergency braking
evaluated using the TTC, peak value of braking (MAEB) based on in-depth crashes in Australia,” Journal of
Crash Prevention and Injury Control, vol. 17, no. 1, pp. 66–72,
deceleration, distance after braking, relative distance
2016.
between braking and stopping, success rate of sensor [8] G. Savino, M. Pierini, J. Thompson, M. Fitzharris, and
recognition, and success rate of collision avoidance. M. G. Lenné, “Exploratory field trial of motorcycle autono-
(4) Currently, the research is only restricted to the mous emergency braking (MAEB): considerations on the
performance of safety from the individual vehicle acceptability of unexpected automatic decelerations,” Traffic
perspective, without considering other performance Injury Prevention, vol. 17, no. 8, pp. 855–862, 2016.
metrics (e.g., environmental footprint) and the im- [9] G. Savino, M. Rizzi, and J. Brown, “Further development of
pact of the AEB system at the system level. The motorcycle autonomous emergency braking (maeb), what can
in-depth studies tell us? A multinational study,” Journal of
change and impact on vehicle stability, energy
Crash Prevention and Injury Control, vol. 15, no. 1,
consumption, and overall traffic efficiency after
pp. S165–S172, 2014.
equipping AEB system (to some extent) would be- [10] M. Ariyanto, G. Haryadi, and M. Munadi, “Development of
come a research direction in the future. low-cost autonomous emergency braking system (AEBS) for
an electric car,” in Proceedings of the 2018 5th International
Data Availability Conference on Electric Vehicular Technology (ICEVT), Octo-
ber, 2018.
No data were used in this study. [11] Y. Gu, Research on Longitudinal Collision Warning and
Automatic Emergency Brake Controller of Electric Vehicle,
Conflicts of Interest Beijing Jiaotong University, Beijing, China, 2018.
[12] W. Jiang and B. Liu, “Modeling and simulation of big bus
The authors declare that they have no conflicts of interest. equipped with autonomous emergency braking system,” in
Proceedings of the 2016 7th International Conference on Mecha-
tronics, Control and Materials (ICMCM 2016), January, 2016.
Acknowledgments [13] J. Fan, “Development of automatic emergency braking system
for special vehicles,” Automobile Applied Technology, vol. 24,
This research has been supported by the National Key Research
pp. 146-147, 2019.
and Development Program of China (No. 2021YFB2501200), [14] J. Guo, P. Zhao, and J. Fan, “Research on control strategy of
the National Natural Science Foundation of China (No. automatic emergency braking for special vehicles,” Vehicle &
61703053), the Shaanxi Province Key Research and Develop- Power Technology, vol. 157, no. 01, pp. 13–17, 2020.
ment Program (2022GY-300), the Shaanxi Province Post- [15] M. Schratter, S. Amler, and P. Daman, “Optimization of the
doctoral Science Foundation (No. 2018BSHYDZZ64), and braking strategy for an emergency braking system by the
Chang'an University Graduate Student Innovation Ability application of machine learning,” in Proceedings of the 2018
Cultivation Project (No. 300203211242). IEEE Intelligent Vehicles Symposium (IV), June, 2018.
[16] X. He, X. Ji, and K. Yang, “AUtonomous emergency braking
References based on radial basis function neural network variable
structure control for collision avoidance,” in Proceedings of
[1] WHO, Global Status Report on Road Safety, World Health the 2017 IEEE 2nd Information Technology, Networking,
Organization, Geneva, Switzerland, 2015. Electronic and Automation Control Conference (ITNEC),
[2] W. Yang, X. Zhang, and Q. Lei, “Research on longitudinal IEEE, Chengdu, China, December, 2017.
active collision avoidance of autonomous emergency braking [17] L. Wang, Z. Zhan, and Y. Xin, “Development of BP neural
pedestrian system (AEB-P),” Sensors, vol. 19-21, 2019. network PID controller and its application on autonomous
[3] H Jeppsson and N Lubbe, “Simulating automated emergency emergency braking system,” in Proceedings of the 2018 IEEE
braking with and without torricelli vacuum emergency Intelligent Vehicles Symposium (IV), June, 2018.
braking for cyclists: effect of brake deceleration and sensor [18] T. Bo, S. Chien, and H. Zhi, “Pedestrian protection using the
field-of-view on accidents, injuries and fatalities,” Accident integration of V2V and the pedestrian automatic emergency
Analysis & Prevention, vol. 142, Article ID 105538, 2020. braking system,” in Proceedings of the 2016 IEEE 19th In-
[4] H. Chajmowicz, J. Saadé, and S. Cuny, “Prospective assess- ternational Conference on Intelligent Transportation Systems
ment of the effectiveness of autonomous emergency braking (ITSC), IEEE, Rio de Janeiro, Brazil, November, 2016.
in car-to-cyclist accidents in France,” Traffic Injury Preven- [19] J. B. Cicchino, “Effectiveness of forward collision warning and
tion, vol. 20, no. 2, pp. S20–S25, 2019. autonomous emergency braking systems in reducing front-to-
[5] G. Savino, J. Brown, and M. Rizzi, “Triggering algorithm rear crash rates,” Accident Analysis & Prevention, vol. 99,
based on inevitable collision states for autonomous emer- pp. 142–152, 2017.
gency braking (AEB) in motorcycle-to-car crash recon- [20] E. Jeong and C. Oh, “Methodology for estimating safety
structions,” in Proceedings of the IEEE Intelligent Vehicles benefits of advanced driver assistant systems,” The Journal of
Symposium, Seoul, Korea, July, 2015. The Korea Institute of Intelligent Transport Systems, vol. 12,
[6] G. Savino, F. Giovannini, and N. Baldanzini, “Assessing the no. 3, pp. 6–77, 2013.
potential benefits of the motorcycle autonomous emergency [21] B. Fildes, M. Keall, N. Bos et al., “Effectiveness of low speed
braking using detailed crash reconstructions,” Journal of autonomous emergency braking in real-world rear-end
12 Journal of Advanced Transportation

crashes,” Accident Analysis & Prevention, vol. 81, pp. 24–29, Automation Science & Engineering IEEE, New Taipei, Taiwan,
2015. August, 2014.
[22] I. Isaksson-Hellman and M. Lindman, “Evaluation of the [39] I. Koglbauer, J. Holzinger, A. Eichberger, and C. Lex, “Au-
crash mitigation effect of low-speed automated emergency tonomous emergency braking systems adapted to snowy road
braking systems based on insurance claims data,” Traffic conditions improve drivers’ perceived safety and trust,” Traffic
Injury Prevention, vol. 17, no. sup1, pp. 42–47, 2016. Injury Prevention, vol. 19, no. 3, pp. 332–337, 2018.
[23] F. Liu, P. Wang, and J. Chen, “Research on the AEB system [40] M. Lin, J. Yoon, and B. Kim, “Proposal and validation of AEB
parameters based on passenger car-pedestrian accidents in system Algorithm for various slope environments,” Advanced
CIDAS,” Research and Application, vol. 9, pp. 26–29, 2018. Multimedia and Ubiquitous, Springer, Berlin Heidelberg,
[24] J. Lenard, R. Welsh, and R. Danton, “Time-to-Collision Germany, 2016.
analysis of pedestrian and pedal-cycle accidents for the de- [41] H. Bae, J. Ryu, and J. Gerdes, “Road Grade and Vehicle
velopment of autonomous emergency braking systems,” Parameter Estimation for Longitudinal Control Using GPS,”
Accident Analysis & Prevention, vol. 115, no. Jun, pp. 128–136, 2001.
2018. [42] G. Kim, H. Mun, and B. Kim, “Performance of AEB system on
[25] Y. Zhao, D. Ito, and K. Mizuno, “AEB effectiveness evaluation a slope using an extended kalman filter,” International Journal
based on car-to-cyclist accident reconstructions using video of Software Engineering and Knowledge Engineering, vol. 29,
of drive recorder,” Traffic Injury Prevention, vol. 20, no. 1, no. 7, pp. 955–969, 2019.
pp. 100–106, 2019. [43] A. Vahidi, A. Stefanopoulou, and H. Peng, “Recursive least
[26] J. Stellet, P. Vogt, and J. Schumacher, “Analytical derivation of squares with forgetting for online estimation of vehicle mass
performance bounds of autonomous emergency brake sys- and road grade: theory and experiments,” Vehicle System
tems,” Intelligent Vehicles Symposium IEEE, , Gothenburg, Dynamics, vol. 43, no. 1, pp. 31–55, 2005.
Sweden, June, 2016. [44] H.-K. Lee, S.-G. Shin, and D.-S. Kwon, “Design of emergency
[27] J. Chen, “Exploration and analysison maintenance method braking algorithm for pedestrian protection based on multi-
and skills of vehicle brake delay fault,” Automobile Applied sensor fusion,” International Journal of Automotive Tech-
Technology, vol. 23, no. 24, pp. 229-230, 2019. nology, vol. 18, no. 6, pp. 1067–1076, 2017.
[28] Y. Hu, Z. Lv, and X. Liu, “The longitudinal collision avoidance [45] H. Cho, Y. Seo, B. Kumar et al., “A multi-sensor fusion system
algorithm and simulation verification of AEB system based on for moving object detection and tracking in urban driving
PreScan,” Journal of Automobile Safety and Energy Conser- environments,” in Proceedings of the IEEE International
vation, vol. 8, no. 2, pp. 136–142, 2017.
Conference on Robotics & Automation IEEE, Hong Kong,
[29] L. Gu and T. Sun, “Simulation comparison of AEB system
China, May, 2014.
with different control strategies in PreScan,” Agricultural
[46] R. Chavez-Garcia, J. Burlet, and T. Vu, “Frontal object per-
Equipment & Vehicle Engineering, vol. 56, no. 10, pp. 76–79,
ception using radar and mono-vision,” in Proceedings of the
2018.
Intelligent Vehicles Symposium IEEE, Madrid, Spain, June,
[30] L. Yi, R. Milton, and L. Kiliaris, “Driver’s style classification
2012.
using jerk analysis,” in Proceedings of the 2009. CIVVS ’09.
[47] C. Premebida, O. Ludwig, and U. Nunes, “LIDAR and vision-
IEEE Workshop on IEEE, Nashville, TN, USA, March, 2009.
based pedestrian detection system,” Journal of Field Robotics,
[31] A. Pentland and A. Liu, “Modeling and prediction of human
vol. 26, no. 9, pp. 696–711, 2010.
behavior,” Neural Computation, vol. 11, no. 1, pp. 229–242,
[48] H. Mun and B. Kim, “A study on the V2V-Communication-
1999.
[32] G. Aoude and G. How, “Using support vector machines and Based AEB system for high-speed driving under various road
bayesian filtering for classifying agent intentions at road in- surface conditions,” Advanced Multimedia and Ubiquitous
tersections,” 2009. Engineering, 2016, Beijing, China, April.
[33] S. Jiang, H. Yang, and T. Wu, “Autonomous emergency [49] M. Liu, S. Chien, and Y. Chen, “Improve road safety using
braking control strategy based on driving state recognition,” combined V2V and pre-collision systems,” in Proceedings of
Automobile Technology, vol. 10, pp. 53–57, 2019. the Geoscience & Remote Sensing, Igarss 97 Remote Sensing-A
[34] R. Miller and Q. Huang, “An adaptive peer-to-peer collision Scientific Vision for Sustainable Development, IEEE Interna-
warning system,” in Proceedings of the IEEE Vehicular tional IEEE, Singapore, June, 2015.
Technology Conference IEEE, Birmingham, AL, USA, May, [50] O. Op den Camp, S. van Montfort, J. Uittenbogaard, and
2002. J. Welten, “Cyclist target and test setup for evaluation of
[35] R. Rajamani, N. Piyabongkarn, and J. Lew, “Tire-road fric- cyclist-autonomous emergency braking,” International Jour-
tion-coefficient estimation,” IEEE Control Systems, vol. 30, nal of Automotive Technology, vol. 18, no. 6, pp. 1085–1097,
pp. 54–69, 2010. 2017.
[36] Y. Hwang and S. B. Choi, “Adaptive collision avoidance using [51] S. E. Li, G. Li, J. Yu et al., “Kalman filter-based tracking of
road friction information,” IEEE Transactions on Intelligent moving objects using linear ultrasonic sensor array for road
Transportation Systems, vol. 20, no. 1, pp. 348–361, 2019. vehicles,” Mechanical Systems and Signal Processing, vol. 98,
[37] A. Sevil, M. Canevi, and M. Soylemez, “Development of an no. 1, pp. 173–189, 2018.
adaptive autonomous emergency braking system based on [52] G. Li, S. E. Li, R. Zou, Y. Liao, and B. Cheng, “Detection of
road friction,” in Proceedings of the 11th International Con- road traffic participants using cost-effective arrayed ultrasonic
ference on Electrical and Electronics Engineering, Bursa, sensors in low-speed traffic situations,” Mechanical Systems
Turkey, November, 2019. and Signal Processing, vol. 132, pp. 535–545, 2019.
[38] I. Han, B. Luan, and F. Hsieh, “Development of autonomous [53] W. Song, Y. Yi, and M. Fu, “Lane detection and classification
emergency braking control system based on road friction,” in for forward collision warning system based on stereo vision,”
Proceedings of the IEEE International Conference on IEEE Sensors Journal, vol. 18, 2018.
Journal of Advanced Transportation 13

[54] M. Park, S. Lee, and C. Kwon, “Design of pedestrian target obstacle collision warning system based on deceleration for
selection with funnel map for pedestrian AEB system,” IEEE collision avoidance,” Review of Automotive Engineering,
Transactions on Vehicular Technology, vol. 66, 2017. vol. 30, 2009.
[55] A. Polychronopoulos, M. Tsogas, and A. Amditis, “Dynamic [70] K. Morita, M. Sekine, and T. Okada, “Factors with the greatest
situation and threat assessment for collision warning systems: influence on drivers’ judgment of when to apply brakes,” in
the EUCLIDE approach,” in Proceedings of the IEEE Intelli- Proceedings of the SICE-ICASE, 2006. International Joint
gent Vehicles Symposium, IEEE, Parma, Italy, Jnue, 2004. Conference IEEE, usan, Korea (South), October, 2006.
[56] J. Hillenbrand, A. M. Spieker, and K. Kroschel, “A multilevel [71] A. Doi, “Development of a rear-end collision avoidance
collision mitigation Approach&mdash;Its situation assess- system with automatic brake control,” JSAE Review, vol. 15,
ment, decision making, and performance tradeoffs,” IEEE no. 4, pp. 335–340, 1994.
Transactions on Intelligent Transportation Systems, vol. 7, [72] Y. Fujita, “Radar brake system,” JSAE Review, vol. 16, no. 1,
no. 4, pp. 528–540, 2006. p. 113, 1995.
[57] J. Hillenbrand, K. Kroschel, and V. Schmid, “Situation as- [73] A. Burgett, A. Carter, and G. Preziotti, “An algorithm for rear-
sessment algorithm for a collision prevention assistant,” in end collision avoidance warning systems,” 2001.
Proceedings of the IEEE Intelligent Vehicles Symposium IEEE, [74] P. Seiler, B. Song, and J. Hedrick, “Development of a collision
Las Vegas, NV, USA, June, 2005. avoidance system,” Neurosurgery, vol. 46, no. 2, p. 492, 1998.
[58] Y. Yingying Zhang, D. Danya Yao, T. Z. Qiu, fnm Lihui Peng, [75] s. Moon and K. Yi, “Human driving data-based design of a
and fnm Yi Zhang, “Pedestrian safety analysis in mixed traffic vehicle adaptive cruise control algorithm,” Vehicle System
conditions using video data,” IEEE Transactions on Intelligent Dynamics, vol. 46, no. 8, pp. 661–690, 2008.
Transportation Systems, vol. 13, no. 4, pp. 1832–1844, 2012. [76] R. He and H. Peng, “Research and development of autono-
[59] A. Rosado, S. Chien, and L. Li, “Certainty and critical speed mous emergency brake (AEB) technology,” Journal Auto-
for decision making in tests of pedestrian automatic emer- motive Safety and Energy, vol. 10, no. 01, pp. 1–15, 2019.
gency braking systems,” IEEE Transactions on Intelligent [77] F. Schuldt, S. Ulbrich, and T. Menzel, “Defining and substan-
Transportation Systems, vol. 18, no. 99, pp. 1358–1370, 2017. tiating the terms scene, situation, and scenario for automated
[60] T. Shimizu and P. Raksincharoensak, “Motion planning via driving,” in Proceedings of the Intelligent Transportation Systems
optimization of risk quantified by collision velocity accom- (ITSC), Gran Canaria, Spain, October, 2015.
panied with AEB activation,” in Proceedings of the 2017 IEEE [78] W.-H. Ma and H. Peng, “A worst-case evaluation method for
International Conference on Vehicular Electronics and Safety dynamic systems,” Journal of Dynamic Systems, Measurement,
(ICVES), June, 2017. and Control, vol. 121, no. 2, pp. 191–199, 1999.
[61] S. Cafiso, A. Garcia, and R. Cavarra, “Pedestrian Crossing [79] D. Zhao, H. Peng, and S. Bao, “Accelerated evaluation of
Safety Improvements: Before and after Study Using Traffic automated vehicles using extracted naturalistic driving data,”
Conflict Techniques,” in Proceedings of the 4th International The Dynamics of Vehicles on Roads and Tracks, CRC Press,
Symposium on Highway Geometric Design, Valencia, Spain, Boca Raton, Florida, 2016.
June, 2010. [80] D. Zhao, H. Lam, and H. Peng, “Accelerated evaluation of
[62] Y. Saito and P. Raksincharoensak, “Shared control in risk automated vehicles safety in lane-change scenarios based on
predictive braking maneuver for preventing collisions with importance sampling techniques,” IEEE Transactions on Intel-
pedestrians,” IEEE Transactions on Intelligent Vehicles, vol. 1, ligent Transportation Systems, vol. 18, no. 3, pp. 595–607, 2017.
pp. 314–324, 2017. [81] Z. Huang, M. Arief, and H. Lam, “Synthesis of different
[63] M. Abu, Z. Kornain, and I. Iqbal, “Automated car braking autonomous vehicles test approaches,” in Proceedings of the
system: using neural network system via labview environ- 2018 21st International Conference on Intelligent Trans-
ment,” in Proceedings of the Open Systems IEEE, Kuala portation Systems (ITSC), November, 2018.
Lumpur, Malaysia, October, 2013. [82] D. Fleury and T. Brenac, “Accident prototypical scenarios, a
[64] S. Christopoulos, S. Kanarachos, and A. Chroneos, “Learning tool for road safety research and diagnostic studies,” Accident
driver braking behavior using smartphones, neural networks Analysis & Prevention, vol. 33, no. 2, pp. 267–276, 2001.
and the sliding correlation coefficient: road anomaly case [83] S. Huang, J. Yang, and F. Eklund, “Evaluation of remote
study,” IEEE Transactions on Intelligent Transportation Sys- pedestrian sensor system based on the analysis of car-pe-
destrian accident scenarios,” Safety Science, vol. 46, no. 9,
tems, vol. 20, no. 1, pp. 65–74, 2018.
pp. 1345–1355, 2008.
[65] D. R. Llorca, M. A. Sotelo, I. Parra, J. E. Naranjo, M. Gavilan,
[84] W. Niewöhner, F. Roth, and J. Gwenenberger, “Proposal for a
and S. Alvarez, “An experimental study on pitch compen-
Test Procedure of Assistance Systems Regarding Preventive
sation in pedestrian-protection systems for collision avoid-
Pedestrian Protection,” 2011.
ance and mitigation,” IEEE Transactions on Intelligent
[85] J. Lenard and R. Danton, “Accident Data Study in Support of
Transportation Systems, vol. 10, no. 3, pp. 469–474, 2009.
Development of Autonomous Emergency Braking Test Pro-
[66] C. Choi and Y. Kang, “Simultaneous braking and steering control
cedures,” LUEL 5989/6175, 2010.
method based on nonlinear model predictive control for
[86] U. Sander and N. Lubbe, “The potential of clustering methods to
emergency driving support,” International Journal of Control,
define intersection test scenarios: assessing real-life performance
Automation and Systems, vol. 15, no. 1, pp. 345–353, 2017.
of AEB,” Accident Analysis & Prevention, vol. 113, pp. 1–11, 2018.
[67] D. N. Lee, “A theory of visual control of braking based on
[87] Q. Chen, X. Lian, and X. Li, “Verify the effectiveness of the
information about time-to-collision,” Perception, vol. 5, no. 4,
AEB pedestrian system through real traffic accident data,”
pp. 437–459, 1976.
Inspection & Identification, vol. 4, pp. 78–80, 2018.
[68] V. Milanés, J. Pérez, J. Godoy, and E. Onieva, “A fuzzy aid
[88] X. Xu, Z. Zhou, and W. Hu, “Intersection test scenarios for
rear-end collision warning/avoidance system,” Expert Systems
AEB based on accident data mining,” Journal of Beijing
with Applications, vol. 39, no. 10, pp. 9097–9107, 2012.
University of Aeronautics and Astronautics, vol. 46, no. 10,
[69] T. Hiraoka, “Collision risk evaluation index based on de-
pp. 4–12, 2020.
celeration for collision avoidance (second report) - forward