DOI: https://doi.org/10.21203/rs.3.rs-2286561/v1
With the continuous improvement of people's living standards, the number of cars has also increased dramatically. While cars are convenient for people to travel, they also lead to increasingly serious traffic safety problems. For this reason, this paper uses the fault tree and Bayesian network methods to conduct an in-depth study on the causes of pedestrian-vehicle traffic accidents from three aspects: people, vehicle, road and the environment. In this paper, the occurrence of pedestrian-vehicle traffic accidents is divided into 29 basic events. The basic events of each of the 381 pedestrian-vehicle traffic accidents were Classified by 0–1. A fault tree model leading to pedestrian-vehicle traffic accidents is established, which is then transformed into a Bayesian network model, and Bayesian network inference, sensitivity analysis is performed with the help of Netica software. Our results suggest that illegal crossing of traffic lanes, speeding, rainy day, slippery road, braking is not timely, visual impairment are the main causes of pedestrian-vehicle traffic accidents. These results can not only provide a reference for transportation technology, but also provide a basis for government legislation.
China's private car fleet reaches 262 million by 2021. With the increase in the number of cars, traffic accidents have become a major social safety issue1. Every year, about 1.2 million people die in traffic accidents worldwide, and tens of millions more are injured or disabled in traffic accidents2,3. The number of traffic accidents and their tragic consequences reveal the need to study the causes of traffic accidents. In order to make a breakthrough in the depth study of road traffic accidents, each country has established a vehicle accident depth study database. For example, the German Federal Institute of Transport (BASt) and the US National Highway Traffic Safety Administration (NHTSA) have established in-depth road traffic accident research systems and databases respectively. China established the China Vehicle Accident In-Depth Investigation System NAIS (National Automobile Accident In-Depth Investigation System) in 2011 by the Defective Products Management Center of the General Administration of Quality Supervision, Inspection and Quarantine of China. In 2020, there were 244674 traffic accidents in China, with 1349 pedestrians killed in accidents and direct economic losses of nearly 30 million yuan. According to the analysis of the NAIS database on the in-depth survey data of vehicle accidents in 2020, the number of pedestrian-vehicle traffic accidents ranked second only to automobiles and two/three-wheelers among all types of traffic accidents. Pedestrians have received a great deal of attention as a vulnerable group of road users4. Especially with the progress of intelligent driving technology, the research on the causes of pedestrian-vehicle traffic accidents can not only provide a reference for traffic technology, but also provide a basis for government legislation, etc.
Regarding the causal factors of traffic accidents, on the study of human factors, Topolšek5 obtained that Wrong-way driving is one of the causes of accidents on highways; Komada6 found drowsiness as a significant factor in human error leading to traffic accidents; Li7 identified not giving way, illegal reversing, illegal parking, drunk driving, and speeding as high-risk behaviors; Wang8 Analyzed the severity of injuries in traffic accidents caused by driver fatigue. On the study of vehicle factors, Vranješ9 obtained that the main vehicle factors leading to traffic accidents are the malfunctioning of lights or light signal devices. On the study of road and environmental factors, Kurdin10 analyzed the influence of geometry and road environment on road accidents in Kecamatan Abeli; Liu11 found that severe weather has a significant impact on the consequences of high-speed accidents. In addition, Angin and Ali12 found that negligent driving and speeding are the main causes of road traffic accidents through a study of people, vehicle, road and the environment; Hauer13 studied the relationship between average annual daily traffic volume, number of commercial lanes, and speed limits and the probability of traffic accidents; Hou14 conducted a detailed inference analysis based on traffic characteristics, highway geometry, pavement conditions, and weather conditions to derive the main causal factors of highway vehicle crashes; Li and Liu15 identified time of day, weather, number of patrol vehicles and surveillance, and age and attributes of the accident driver as some of the factors contributing to high speed traffic accidents; Traffic accidents are the result of a combination of factors, human, vehicle, road and the environment16. In this paper, a total of 29 basic pedestrian-vehicle traffic accident causative factors, including drunk driving and pedestrians crossing the traffic lane illegally, are also obtained from three categories: people, vehicle, road and environment.
Currently, there are various methods used to study the causes of traffic accidents. For example, Keay and Simmonds17 designed a regression model for the effect of weather on traffic; Yu18 established a highway collision model using Poisson model and Bayesian inference method. Selvasofia and Arulraj19 used GIS for traffic accident analysis; Zamzuri20 used Bayesian network and HC algorithm and Tabu algorithm to explore the causes of traffic accidents in Malaysia; Zhang21 used text processing technology based on LDA Topic Model to analyze traffic accidents to obtain the most dominant factors of traffic accidents; Deng22 proposed a causality analysis model for traffic accidents using a hybrid AHP and Apriori-Gentic based algorithm to mine accident causes. Bayesian network is a good way to model causality, and Bayesian network can effectively form probabilistic models for efficient inference and learning23. Fault tree analysis is an important analysis method for system reliability and safety analysis24. In this paper, we combine fault tree and Bayesian network methods to build a model of pedestrian-vehicle traffic accidents, which can more accurately identify the factors affecting pedestrian-vehicle traffic accidents. We first construct a fault tree model for pedestrian-vehicle traffic accidents, and then transform the fault tree model into a Bayesian network model, which can generate the relationship structure between variables. Bayesian network can quantify the relationships between variables as probabilities, and using these probability values can make it relatively easy to explore the relationships between variables and make inference about pedestrian-vehicle traffic accidents25.
The main objective of this paper is to analyze the main causal factors leading to pedestrian-vehicle traffic accidents using fault tree and Bayesian network models. Provide guidance for the traffic safety department to formulate corresponding preventive measures, reduce the number of traffic accidents of people and vehicles, and reduce the number of deaths and injuries.
Taking 2017–2019 as an example, 381 pedestrian-vehicle traffic accidents were collected from the NAIS database. These include 57 minor traffic accidents, 144 general traffic accidents, 138 major traffic accidents, and 42 very large traffic accidents. Information collected includes weather conditions, vehicle conditions, major errors of personnel, etc. The causes of pedestrian-vehicle traffic accidents are complex and varied, but overall they are broadly divided into three categories: people, vehicle, road and the environment. Therefore, the information collected is sorted into these three main categories. The factors that we want to obtain to cause pedestrian-vehicle traffic accidents are binary events, where 1 means that the event occurs and 0 means that it does not occur. In accordance with this principle, take a 2018 pedestrian-vehicle traffic accident as an example, in which a certain person drove a Hyundai car on a city road after drinking on March 21 and collided with a pedestrian crossing the highway, resulting in serious injuries to the pedestrian. At this time, the drunk driving of the motor vehicle driver and the illegal crossing of the traffic lane by pedestrian are recorded as 1, and the other basic events are recorded as 0. By classifying 0–1 basic events of 381 pedestrian-vehicle traffic accidents one by one, and recording the frequency of occurrence of each basic event, the following data are obtained. As shown in Table 1.
| Base event name | Frequency | Base event name | Frequency |
|---|---|---|---|
| Illegal crossing of traffic lanes | 201 | Early disease | 6 |
| Illegal crossing of isolation facilities | 27 | Fatigue driving | 15 |
| Violation of traffic signals | 21 | Flat tire | 6 |
| Traffic environment not observed | 42 | Steering failure | 6 |
| Driver violates traffic signals | 9 | Brake failure | 9 |
| Speeding | 108 | Speeding | 108 |
| Wrong-way driving | 6 | Overload | 45 |
| Drunk driving | 48 | Poor road condition | 24 |
| Illegal overtaking | 6 | Visual impairment | 69 |
| Illegal U-turn | 6 | Slippery road | 93 |
| Not keeping a safe distance | 42 | Rainy day | 99 |
| Not giving way | 33 | Snow day | 9 |
| Braking is not timely | 90 | Foggy day | 42 |
| Misjudgment in an emergency | 9 | Illegal lane-changing | 6 |
| Mistakes in Emergency Situations | 9 |
A fault tree model is a logical causal relationship diagram consisting of top events, intermediate events, bottom events and logical symbols. The logical symbols also include three logical relationships of and, or, not. The basic symbol of fault tree is shown in Fig. 1.
The Bayesian formula originally originated from a paper published by the British scholar Bayes, which is the theoretical basis for inference and calculation of Bayesian network26. If there are events such as \({A}_{1},{A}_{2},\cdots ,{A}_{n}\), etc. they are mutually exclusive and constitute a complete event. And \(p({A}_{i}⩾0 (i=\text{1,2},\cdots ,n)\), event \(C\) and \({A}_{1},{A}_{2},\cdots ,{A}_{n}\) occur at the same time, then the Bayesian formula can be expressed as:
Bayesian network, also known as Belief Network, is a conceptual model proposed by Pearl in 198827. BN is considered to be one of the most effective theoretical models in the field of uncertain knowledge representation and reasoning, and its topology is Directed Acyclic Graphical (DAG) 28. Nodes in BN represent random variables, and directed edges represent causal relationships or unconditional independence between variables. The source point of the arrow is the parent node, the arrow points to the child node, the root node has no parent node, and the leaf node has no child node.
Each node has a Conditional Probability Table (CPT), which represents the relationship with all its parent nodes, including prior and posterior probabilities29. The prior probability is the probability distribution of a node without a parent node; The posterior probability can be obtained by modifying the prior probability according to the actual information. The Equation for calculating the prior probability of the root node is as follows:
In Eq. (2), \(n\) is the frequency of events at the root node of the node, and m is the number of pedestrian-vehicle traffic accidents collected.
Let BN be \(U=\{{X}_{1},{X}_{2},\cdots ,{X}_{n}\}\), \({X}_{i}\) represents a node, where \({X}_{1},{X}_{2},\cdots ,{X}_{n}\) are \(n\) discrete variables. All parent nodes of \({X}_{i}\) are represented by \(\text{P}\text{a}\text{r}\left({X}_{i}\right)\), and non-descendant nodes are represented by \(A\left({X}_{i}\right)\). In BN, under the premise that \(\text{P}\text{a}\text{r}\left({X}_{i}\right)\) is known, \(A\left({X}_{i}\right)\) and \({X}_{i}\) are conditionally independent,which is:
The interdependence between nodes is defined by the joint probability distribution, and the joint probability distribution of all nodes can be expressed as:30
To convert a fault into a corresponding Bayesian network, it is necessary to establish a correspondence between the two: event-node, logic gate-connection strength. The following are the specific conversion steps: (1) The bottom event in the fault tree is transformed into the root node in the BN, the intermediate event in the fault tree is transformed into the intermediate node in the BN, and the top event in the fault tree is transformed into the leaf node in the BN. (2) The probability value of the occurrence of the bottom event in the fault tree corresponds to the prior probability value of the root node in the BN. (3) The input and output relationship between the logic gate in the fault tree and the events above and below the logic gate should correspond to the arrow direction of the directed connection between the nodes in the BN. (4) Transform the relationship between the logic gates in the fault tree into the conditional probability of the nodes in the BN. For example, the following Fig. 2 is a simple AND gate and OR gate event, and its logical relationship can be represented by the conditional probability of the nodes in BN. Among them, the M, N, and Q events in the fault tree are all binary events, 1 indicates that the event occurs, and 0 indicates that the event does not occur. The fault tree is transformed into a Bayesian network algorithm as shown in the Fig. 2.
First, determine the top event: Take the pedestrian-vehicle traffic accident as the top event.
Second, determine intermediate events: take human, vehicle, road and environment as the first layer of intermediate events. human include motor vehicle drivers and pedestrians. Motor vehicle drivers are further divided into bad state and misconduct. Bad state refers to the occurrence of pedestrian-vehicle traffic accidents due to virtual objective conditions such as emotions and states of drivers. Misconduct refers to the driver's subjective violation of traffic laws or operational errors that lead to the occurrence of pedestrian-vehicle traffic accidents. Vehicle include faulty vehicles and safety hazards. road and environment include factors such as bad weather.
Finally, determine the bottom events: Initially, through discussions with expert, it was found that the bad states included distracted driving, emotional driving, fatigue driving, etc. Mistakes include illusion, braking is not timely, not giving way, etc. Violations include speeding, driving without a license, drunk driving, illegal lane-changing, etc. Pedestrians include illegal crossing of traffic lanes, illegal crossing of isolation facilities, violation of traffic signals, and traffic environment not observed. However, distracted driving, emotional driving, and illusion were not recorded in the NAIS database, and the probability of driving without a license is almost 0. Therefore, these situations are not typical, and after discussion and reflection with expert, it was decided not to join these events. The bottom events of the vehicle, road and environment will not be repeated, as shown in the Fig. 3.
Combined with the above analysis of each layer of the fault tree, the logical relationship between events is determined. The events of each layer are connected in series from top to bottom by using logic gates to form a pedestrian-vehicle traffic accident. The constructed fault tree is shown in the Fig. 3.
The establishment of the Bayesian network topology must first determine each node. We can convert the fault tree model above into a Bayesian network topology according to the steps of transforming the fault tree into a Bayesian network. Use Netica software to draw the model and input the prior probability of the root node and the conditional probability table of the non-root node, as shown in the Fig. 4.
First, determine the prior probability of the root node. Since the fault tree event is binary, the nodes of the Bayesian network model transformed from the fault tree model are also binary, 1 means the node event occurs, and 0 means the node event does not occur. The frequency that the root node state is 1 is the probability value of the event, and the probability value of the root node event is calculated according to Eq. (2). According to the complementarity of events, the probability value of the state of each root node being 0 can be obtained. The prior probability of the root node is shown in Table 2.
| Root node | P(Xi=1) | P(Xi=0) | Root node | P(Xi=1) | P(Xi=0) |
|---|---|---|---|---|---|
| X1 | 0.528 | 0.472 | X16 | 0.016 | 0.984 |
| X2 | 0.071 | 0.929 | X17 | 0.039 | 0.961 |
| X3 | 0.055 | 0.945 | X18 | 0.016 | 0.984 |
| X4 | 0.110 | 0.890 | X19 | 0.016 | 0.984 |
| X5 | 0.024 | 0.976 | X20 | 0.024 | 0.976 |
| X6 | 0.283 | 0.717 | X21 | 0.283 | 0.717 |
| X7 | 0.016 | 0.984 | X22 | 0.118 | 0.882 |
| X8 | 0.126 | 0.874 | X23 | 0.063 | 0.937 |
| X9 | 0.016 | 0.984 | X24 | 0.181 | 0.819 |
| X10 | 0.016 | 0.984 | X25 | 0.244 | 0.756 |
| X11 | 0.110 | 0.890 | X26 | 0.260 | 0.740 |
| X12 | 0.087 | 0.913 | X27 | 0.024 | 0.976 |
| X13 | 0.236 | 0.764 | X28 | 0.110 | 0.890 |
| X14 | 0.024 | 0.976 | X29 | 0.016 | 0.984 |
| X15 | 0.024 | 0.976 |
Second, determine conditional probability Table of non-root nodes. The fault tree shown in the Figure 2 is transformed into a Bayesian network algorithm, and the CPT of the non-root node can be obtained. Taking the A7 node as an example, the corresponding CPT is shown in Table 3, and the rest of the non-root nodes are similar.
| Parent node status | Node A7 Conditional Probability | |||
|---|---|---|---|---|
| X18 | X19 | X20 | P(A7 = 1) | P(A7 = 0) |
| 0 | 0 | 0 | 0 | 1 |
| 0 | 0 | 1 | 1 | 0 |
| 0 | 1 | 0 | 1 | 0 |
| 0 | 1 | 1 | 1 | 0 |
| 1 | 0 | 0 | 1 | 0 |
| 1 | 0 | 1 | 1 | 0 |
| 1 | 1 | 0 | 1 | 0 |
| 1 | 1 | 1 | 1 | 0 |
The causes of pedestrian-vehicle traffic accidents are mainly considered from three categories: human, vehicle, road and the environment. It can be seen from the Fig. 4 that the probability of the occurrence of pedestrian-vehicle traffic accidents is 97.4%, which is close to 1, indicating that the established Bayesian network is reasonable and feasible. It can also be seen from the Fig. 4 that the probability of pedestrian-vehicle traffic accidents caused by human factors is 88.2%, and the probabilities of vehicle factors, road and environment causing traffic accidents are 40.2% and 62.7%, respectively.
The diagnostic reasoning of Bayesian network can calculate the probability distribution of each root node when a pedestrian-vehicle traffic accident has occurred. Sorting by the size of the probability change value can analyze the importance of each root node when a pedestrian-vehicle traffic accident occurs. The specific steps are as follows, using the probability update of the Bayesian network to take pedestrian-vehicle traffic accidents as the target node. In the Netica software, the probability of the leaf node of the pedestrian-vehicle traffic accidents is set to 100%, and the posterior probability value of each root node is obtained. The prior probability and posterior probability of the root node are combined for analysis, and the importance is sorted according to the size of the probability increase value. The analysis results are shown in Table 4.
| Root node number | Priori probability (%) | Posterior probabilit (%) | Order of importance | Root node number | Priori probability (%) | Posterior probabilit (%) | Order of importance |
|---|---|---|---|---|---|---|---|
| X1 | 52.8 | 54.2 | 1 | X16 | 1.6 | 1.64 | 13 |
| X2 | 7.1 | 7.29 | 8 | X17 | 3.9 | 4.01 | 11 |
| X3 | 5.5 | 5.65 | 10 | X18 | 1.6 | 1.64 | 13 |
| X4 | 11 | 11.3 | 6 | X19 | 1.6 | 1.64 | 13 |
| X5 | 2.4 | 2.46 | 12 | X20 | 2.4 | 2.46 | 12 |
| X6 | 28.3 | 29.1 | 2 | X21 | 28.3 | 29.1 | 2 |
| X7 | 1.6 | 1.64 | 13 | X22 | 11.8 | 12.1 | 6 |
| X8 | 12.6 | 12.9 | 6 | X23 | 6.3 | 6.47 | 9 |
| X9 | 1.6 | 1.64 | 13 | X24 | 18.1 | 18.6 | 5 |
| X10 | 1.6 | 1.64 | 13 | X25 | 24.4 | 25.1 | 3 |
| X11 | 11 | 11.3 | 6 | X26 | 26 | 26.7 | 3 |
| X12 | 8.7 | 8.94 | 7 | X27 | 2.4 | 2.46 | 12 |
| X13 | 23.6 | 24.2 | 4 | X28 | 11 | 11.3 | 9 |
| X14 | 2.4 | 2.46 | 12 | X29 | 1.6 | 1.64 | 13 |
| X15 | 2.4 | 2.46 | 12 |
As can be seen from the Table 4, the posterior probability is increased compared to the prior probability. Among them, the top six in order of importance are X1 (illegal crossing of traffic lanes), X6 (speeding), X26 (rainy day), X25 (slippery road), X13 (braking is not timely), X24 (visual impairment). The probability of these six root nodes increases the most, and the risk of causing pedestrian-vehicle traffic accidents is also greater than other root nodes.
In a Bayesian network, the posterior probability of a child node varies with the prior probability of the parent node. Sensitivity analysis quantifies the degree to which the child node changes with the change of the parent node, and obtains the corresponding sensitivity coefficient. In order to find out the factors that have a greater impact on the target node in the Bayesian network of pedestrian-vehicle traffic accidents, and focus on them. And the factors with smaller sensitivity coefficients can be screened out, thereby reducing the complexity of the network structure. Sensitivity analysis can be performed using Netica software. First select the leaf node pedestrian-vehicle traffic accident, and then select the Sensitivity to Findings option for sensitivity analysis. The sensitivity of the remaining nodes to the leaf node is shown in the Table 5.
| Node number | Relevance information | Percentag(%) | Coefficient of variation | Node number | Relevance information | Percentage(%) | Coefficient of variation |
|---|---|---|---|---|---|---|---|
| Q | 0.17574 | 100 | 0.0256545 | X9 | 0.00062 | 0.354 | 0.0000113 |
| M1 | 0.08525 | 48.5 | 0.0051780 | X10 | 0.00062 | 0.354 | 0.0000113 |
| M2 | 0.01991 | 11.3 | 0.0004674 | X11 | 0.00450 | 2.56 | 0.0000860 |
| M3 | 0.03837 | 21.8 | 0.0011674 | X12 | 0.00351 | 2 | 0.0000662 |
| A1 | 0.03881 | 22.1 | 0.0011883 | X13 | 0.01039 | 5.91 | 0.0002145 |
| A2 | 0.04435 | 25.2 | 0.0014714 | X14 | 0.00094 | 0.533 | 0.0000171 |
| A3 | 0.04213 | 24.0 | 0.0013536 | X15 | 0.00094 | 0.533 | 0.0000171 |
| A4 | 0.00215 | 1.23 | 0.0000399 | X16 | 0.00062 | 0.354 | 0.0000113 |
| A5 | 0.02152 | 12.2 | 0.0005165 | X17 | 0.00153 | 0.872 | 0.0000282 |
| A6 | 0.02033 | 11.6 | 0.0004800 | X18 | 0.00062 | 0.354 | 0.0000113 |
| A7 | 0.00218 | 1.24 | 0.0000404 | X19 | 0.00062 | 0.354 | 0.0000113 |
| A8 | 0.01772 | 10.1 | 0.0004036 | X20 | 0.00094 | 0.533 | 0.0000171 |
| A9 | 0.01708 | 9.72 | 0.0003858 | X21 | 0.01285 | 7.31 | 0.0002740 |
| X1 | 0.02912 | 16.6 | 0.0007766 | X22 | 0.00484 | 2.75 | 0.0000929 |
| X2 | 0.00284 | 1.62 | 0.0000531 | X23 | 0.00251 | 1.43 | 0.0000467 |
| X3 | 0.00218 | 1.24 | 0.0000404 | X24 | 0.00770 | 4.38 | 0.0001534 |
| X4 | 0.00449 | 2.56 | 0.0000858 | X25 | 0.01080 | 6.14 | 0.0002241 |
| X5 | 0.00094 | 0.533 | 0.0000171 | X26 | 0.01163 | 6.62 | 0.0002439 |
| X6 | 0.01285 | 7.31 | 0.0002740 | X27 | 0.00094 | 0.533 | 0.0000171 |
| X7 | 0.00062 | 0.354 | 0.0000113 | X28 | 0.00449 | 2.56 | 0.0000858 |
| X8 | 0.00519 | 2.95 | 0.0001001 | X29 | 0.00062 | 0.354 | 0.0000113 |
The correlation information in the Table 5 indicates the degree of dependence of the node on the leaf node. The larger the value, the larger the sensitivity coefficient. Among the secondary indicators, M1 (human) has the largest sensitivity coefficient to pedestrian-vehicle traffic accidents. This is followed by M3(road and environment) and finally M2 (vehicle). In the root node, the top six sensitivity coefficients are X1 (illegal crossing of traffic lanes), X6 (speeding), X26 (rainy day), X25 (slippery road), X13 (braking is not timely), X24 (visual impairment). The bottom seven are X7 (wrong-way driving), X9 (illegal overtaking), X10 (illegal U-turn), X16 (early disease), X18 (flat tire), X19 (steering failure), X29 (illegal lane-changing).
The Bayesian network model is used to analyze the causes of pedestrian-vehicle traffic accidents, and the model analysis results are obtained by combining a posteriori probability inference and sensitivity analysis.
It can be seen from the analysis of the Fig. 5: X1 (illegal crossing of traffic lanes), X6 (speeding), X26 (rainy day), X25 (slippery road), X13 (braking is not timely), X24 (visual impairment) are the main causes of pedestrian-vehicle traffic accidents.
This paper focuses on the pedestrian-vehicle traffic accidents. From the perspectives of human, vehicle, road and the environment, using the fault tree and Bayesian network model, six major causes of pedestrian-vehicle traffic accidents are obtained. They are: X1 (illegal crossing of traffic lanes), X6 (speeding), X26 (rainy day), X25 (slippery road), X13 (braking is not timely), X24 (visual impairment). Therefore, these events should be focused on in daily life.
For these six causative factors, pedestrians crossing traffic lanes illegally is the most important cause of traffic accidents. In order to save time, pedestrians cross the traffic lanes regardless of their own safety, and the driver has no time to respond, which greatly increases the probability of pedestrian-vehicle traffic accidents. For speeding, when speeding on complex road sections and fork roads, the driver cannot observe in time, cannot obtain enough road information, and it is difficult to make a correct judgment on the road conditions. In addition, speeding will shorten the driver's handling time for emergency situations, increase the braking distance of the vehicle, and easily lead to the occurrence of traffic accidents. For rainy day and slippery road, the specific performance is that the driver's field of vision is limited due to rain and cannot obtain road information at a distance. On the other hand, the road surface is slippery, the grip of the vehicle is reduced, the braking performance of the vehicle is reduced, and it is not easy to brake the vehicle in an emergency. For visual impairment, including the limited field of vision of the driver caused by weather, surrounding environment, etc. The inability to obtain sufficient information on the road surface and the environment leads to misjudgment by the driver of the road surface and environmental conditions, resulting in the occurrence of pedestrian-vehicle traffic accidents.
Regarding human factors, many scholars only conduct causal research from the perspective of motor vehicle drivers, ignoring the key role of pedestrians in pedestrian-vehicle traffic accidents. In the analysis of human factors, this paper comprehensively considers the factors of motor vehicle drivers and pedestrians. In our findings, the basic event of pedestrian illegally crossing traffic lanes is the most important pedestrian-vehicle traffic accidents. This proves the importance of adding pedestrian factors.
We should also be aware of some limitations of this study. First, there are many factors that affect pedestrian-vehicle traffic accidents, but this paper only proposes 29 basic events. It is difficult to conduct a comprehensive analysis of the causative factors only by relying on personal literature review and accident data analysis. More accident data should be collected, and the supplementary causal factor indicator system should be further improved through discussions with expert. Second, among the pedestrian-vehicle traffic accidents collected from the NAIS database, there are a few cases where the records of accident information are not very complete due to some reasons.
In summary, this research can better improve traffic operation from the aspects of equipment, facilities, environmental management, education, punishment, etc. So as to better protect the vulnerable group of road users: pedestrians.
Acknowledgements
This study was supported by (1) The Open Research Fund of Sichuan Key Laboratory of Vehicle Measurement, Control and Safety (szjj2018-130); (2) Sichuan Province Innovation Training Project (S202210623048 and S202210623064).
Author contributions
J.Y. conceived and designed the study. Y.L. proposed the method and contributed to the analysis. H.S. and Y.C. made an investigation. J.Y. accumulated resources. J.Y., Y.L. wrote the original draft. J.Y., H.S. and Y.L. contributed to the review and editing. All authors gave their final approval of the manuscript version to be submitted.
Competing interests
The authors declare no competing interests.
Data availability
All data generated or analyzed during this study are included in this published article.