A dangerous event chain extraction method and system based on a complex traffic scene

Abstract

The application discloses a kind of dangerous event chain extraction method and system based on complex traffic scene, including obtaining test scene data;Based on the key factors of dynamic attribute and interaction between vehicle and environment determined by preset natural driving data and sensor performance information, and the uncertainty of key factors is quantified;Based on the risk estimation multidimensional feature set and uncertainty information preset constructs driving risk estimation model for test scene;The driving risk estimation model is optimized, and the optimized risk estimation model for test is obtained;Based on the time evolution characteristics of dangerous event, a dangerous time chain model is constructed according to the optimized risk estimation model;Dangerous time chain model is solved, and dangerous event chain is obtained.The application uses traffic accident data and traffic conflict data to reveal the space-time evolution law of complex traffic scene, and improves the effect of later test scene reproduction and reconstruction by solving the dangerous time chain model.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving control technology, specifically to a method and system for extracting dangerous event chains based on complex traffic scenarios. Background Technology

[0002] Autonomous vehicles are designed to address the safety concerns associated with human driving. However, in real-world driving scenarios, complex elements and strong environmental interference make the driving environment of autonomous driving systems highly uncertain and difficult to repeat or predict. Furthermore, the evolutionary patterns of key spatiotemporal relationships within these scenarios are unclear, and there is a lack of crucial methods for generating test scenarios. High-fidelity scenario generation is challenging, posing a significant challenge to the reproduction and reconstruction of subsequent test scenarios. Summary of the Invention

[0003] The purpose of this invention is to address the problems in the prior art by providing a method for extracting hazard event chains based on complex traffic scenarios. This method can solve the hazard time chain model, thereby improving the effectiveness of scenario reproduction and reconstruction in subsequent testing.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A method for extracting hazard event chains in complex traffic scenarios, characterized by: including:

[0006] S1, Obtain test scenario data;

[0007] S2, based on preset natural driving data and sensor performance information, determine the key factors of dynamic attributes and interactions between the vehicle and the environment, and classify and quantify the uncertainty of the key factors.

[0008] S3, constructs a driving risk estimation model for test scenarios based on a preset multi-dimensional feature set of risk estimation and uncertainty information;

[0009] S4, optimize the driving risk estimation model to obtain an optimized risk estimation model for the test;

[0010] S5. Based on the temporal evolution characteristics of hazardous events, a hazardous time chain model is constructed according to the optimized risk estimation model.

[0011] S6. Solve the dangerous time chain model to obtain the dangerous event chain.

[0012] Preferably, step S1 includes:

[0013] (1) Collect traffic accident data, which includes vehicle, road and environmental data of traffic accidents that occur during the natural driving process of a car;

[0014] (2) Traffic conflict data is collected based on natural driving experiments, including vehicle-to-vehicle conflict data and vehicle-to-road conflict data.

[0015] Preferably, step S2 includes:

[0016] (1) Obtain the uncertainty of the perception result based on the sensor performance information, which includes camera data, millimeter-wave radar data and lidar data;

[0017] (2) Based on the natural driving data, obtain the uncertainty of interaction characteristics and the uncertainty of information blind spots.

[0018] Furthermore, the uncertainty of the perception result includes the results of individual and fused perception and understanding from cameras, millimeter-wave radar, and lidar; the uncertainty of the interaction characteristics includes at least the randomness of traffic participants in mixed traffic; and the uncertainty of the information blind spot includes at least the uncertainty of road conditions, obstruction and absence of traffic facilities.

[0019] Preferably, step S3 includes:

[0020] (1) Construct a multidimensional feature set for risk estimation based on the dynamic attributes and key factors of interaction between the vehicle and the environment determined in step S2;

[0021] (2) Based on the combined effect of each element in the multidimensional feature set of the risk estimation on driving risk, the driving risk estimation model is established. The driving risk estimation model includes the field strength of each traffic element, the force exerted by the risk field on the vehicle under test, and the driving risk coefficient.

[0022] Preferably, step S4 includes:

[0023] (1) Based on simulation software, dangerous accident scenarios were reconstructed and tested to obtain verification results;

[0024] (2) Analyze driving patterns based on the verification results;

[0025] (3) Adjust the weights in the driving risk estimation model according to the driving rules to optimize the driving risk estimation model.

[0026] Furthermore, based on the factors influencing driving risks in the kinetic energy field, potential energy field, and behavioral field, an importance judgment is added, and the weights in the driving risk estimation model are adjusted.

[0027] Preferably, step S5 includes:

[0028] (1) The time series of road traffic accidents and conflicts in the optimized risk estimation model are treated as Markov random processes;

[0029] (2) Based on Markov stochastic processes, a dangerous event chain model is formed by determining the initial probability of different states in the road traffic accident and the transition probability distribution between states on the chain.

[0030] Preferably, step S6 includes:

[0031] (1) Based on the road hazard event status and vehicle driving risk coefficient distribution in the hazard event chain model, the range of the road segment passed through at different times is set to represent the hidden state, and the level is divided;

[0032] (2) Establish a prediction model based on the training dataset, which includes existing road hazard event states;

[0033] (3) Calculate the number of training datasets that are different from the state at the previous time step when they are transitioned to the state at the next time step, and calculate the state transition matrix and the transition probability matrix of the observed variables;

[0034] (4) Determine the initial probability vector based on the state transition matrix, the observed variable transition probability matrix, and the prediction model;

[0035] (5) Determine the predicted value of the hidden state based on the prediction model;

[0036] (6) Predict the state of the actual dangerous event based on the initial probability vector and the hidden state prediction value.

[0037] This invention also provides a system for extracting hazardous event chains based on complex traffic scenarios, comprising:

[0038] The acquisition module is used to acquire test scenario data;

[0039] The quantization module is used to determine the key factors of the dynamic attributes and interactions between the vehicle and the environment based on preset natural driving data and sensor performance information, and to perform hierarchical quantification of the uncertainty of the key factors.

[0040] The estimation model construction module is used to construct a driving risk estimation model for the test scenario based on a preset multidimensional feature set of risk estimation and the uncertainty information.

[0041] The first optimization module is used to optimize the driving risk estimation model to obtain an optimized risk estimation model for the test.

[0042] The second optimization module is used to construct a hazard time chain model based on the temporal evolution characteristics of hazardous events and the optimized risk estimation model.

[0043] The analysis module is used to solve the hazardous time chain model to obtain the hazardous event chain.

[0044] Due to the application of the above technical solutions, the present invention has the following advantages compared with the prior art: The dangerous event chain extraction method based on complex traffic scenarios of the present invention uses the collected traffic accident data and traffic conflict data collected in natural driving experiments to reveal the spatiotemporal evolution law of complex traffic scenarios, study the dangerous event chain extraction method, and improve the effect of subsequent test scenario reproduction and reconstruction by solving the dangerous time chain model. Attached Figure Description

[0045] Figure 1 This is a flowchart of the hazardous event chain extraction method based on complex traffic scenarios in this embodiment;

[0046] Figure 2 This is a schematic diagram of the danger event chain model in this embodiment. Detailed Implementation

[0047] The purpose of this invention is to provide a method and system for extracting hazardous event chains based on complex traffic scenarios. By solving the hazardous time chain model, the effectiveness of subsequent test scenario reproduction and reconstruction is improved.

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] For example, this invention provides a method for extracting hazard event chains based on complex traffic scenarios, the flowchart of which is shown below. Figure 1 As shown, the specific steps include the following.

[0050] Step 100: Obtain test scenario data.

[0051] Specifically, it includes:

[0052] (1) Collect traffic accident data, which includes vehicle, road and environmental data of traffic accidents that occur during the natural driving process of automobiles.

[0053] In this embodiment, traffic accident data is obtained through on-site accident investigation.

[0054] (2) Traffic conflict data were collected based on natural driving experiments. The traffic conflict data included vehicle-to-vehicle conflict data and vehicle-to-road conflict data.

[0055] In this embodiment, a vehicle data acquisition device equipped with sensors collects natural driving data from different drivers, including driving operation information and road environment information.

[0056] Vehicle-to-vehicle conflict data includes lane conflicts, hard braking of the vehicle in front during following, and emergency braking of the vehicle affecting the following vehicle. Lane conflicts include vehicles cutting in the same direction, vehicles occupying the lane in the opposite direction, and changing lanes when there are adjacent vehicles in a natural driving experiment.

[0057] Vehicle-road conflict data specifically includes road construction, road obstacles, road curvature, and road surface dirtiness.

[0058] Step 200: Based on preset natural driving data and sensor performance information, determine the key factors of dynamic attributes and interactions between the vehicle and the environment, and classify and quantify the uncertainty of the key factors.

[0059] Specifically, it includes:

[0060] (1) Obtain the uncertainty of the perception result based on the sensor performance information, which includes camera data, millimeter-wave radar data and lidar data.

[0061] (2) Based on natural driving data, we can obtain uncertainty in interaction characteristics and uncertainty in information blind spots.

[0062] Specifically, the uncertainty of perception results includes the results of individual and fused perception and understanding from sensors such as cameras, millimeter-wave radar, and lidar; the uncertainty of interaction characteristics includes at least the randomness of traffic participants in mixed traffic; and the uncertainty of information blind spots includes at least the uncertainty of road conditions, obstruction or absence of traffic facilities, etc.

[0063] The uncertainty of perception results is mainly determined by the sensor's own performance and environmental conditions, as shown in Table 1. The uncertainty of interaction characteristics is mainly determined by the randomness, commonality, and compliance of participants in mixed traffic, as shown in Table 2, which is a table of uncertainties for traffic participants. The uncertainty of information blind spots is mainly determined by road conditions such as lane line visibility and traffic facilities. The complexity of the road layer is mainly determined by lane line visibility, as shown in Table 3, which is a table of road layer complexity. The complexity of the traffic facility layer is mainly determined by the visibility of traffic facilities, as shown in Table 4, which is a table of traffic facility layer complexity.

[0064] Table 1

[0065] sunny Rainy days, evenings, etc. Ambient lighting at night No ambient light at night Dense fog 1 2 3 4 5

[0066] Table 2

[0067]

[0068] Table 3

[0069]

[0070] Table 4

[0071]

[0072] Step 300: Construct a driving risk estimation model for the test scenario based on the preset multidimensional feature set of risk estimation and uncertainty information.

[0073] In this embodiment, a multi-dimensional feature set for risk estimation is constructed based on the quantified dynamic attributes and key factors of interaction between the vehicle and the environment. By comprehensively analyzing the effects of each element in the multi-dimensional feature set on driving risk, a unified driving risk model is established. This model includes the field strength of each traffic element, the force exerted by the risk field on the tested vehicle, and the driving risk coefficient. Using the field strengths of traffic elements such as people, vehicles, and roads described by the kinetic energy field, potential energy field, and behavioral field, the force exerted by the risk field on the tested vehicle and the driving risk coefficient are calculated.

[0074] The specific implementation process is as follows:

[0075] (1) Based on the combined effect of the human-vehicle-road elements in the surrounding environment on driving risk, a unified driving risk field model is established as shown in equation (1).

[0076] E S =E R +E V +E D (1)

[0077] In the formula, E S For the overall strength of driving risk field, E R For potential energy fields to be strong, E V For a strong kinetic field, E D For behavior field strength.

[0078] (2) By establishing mathematical models of the kinetic energy field, potential energy field and behavioral field formed by various elements in the driving environment, a driving risk field model of the vehicle is constructed, and the total field strength of the driving risk field of the tested vehicle is calculated.

[0079] Among them, the kinetic energy field strength E formed by the moving object a at the location of the tested vehicle j is... V_aj for:

[0080]

[0081] In the formula, r aj =(x j -x a ,y j –y a () represents the distance vector between two points, where k1, k2, and G are undetermined constants greater than zero, and v a Let θ be the velocity of the moving object a. a The direction of velocity is relative to r ajThe included angle, M a Let R be the virtual mass of a moving object a in the road environment. a The road condition factor is the location of the moving object a.

[0082] The potential energy field strength E of the stationary object b at the location of the vehicle j being tested R_bj for:

[0083]

[0084] In the formula, r bj =(x j -x b ,y j –y b () represents the distance vector between two points, k1 and G are undetermined constants greater than zero, and M b Let R be the virtual mass of the stationary object b. b The road condition factor is the location of the stationary object b.

[0085] The field strength E at the location of the tested vehicle j is the behavioral field created by the driver c of surrounding vehicles driving under certain road conditions. D_cj for:

[0086] E D_cj =E V_cj ·D c (4)

[0087] In the formula, E V_cj Let D be the field strength of the kinetic energy field generated by the vehicle driven by driver c at the location of the tested vehicle j. c For driver c, the risk factor.

[0088] Construct a driving risk field model for the vehicle under test, and calculate the total field strength E of the driving risk field for the vehicle under test. S_j :

[0089]

[0090] In the formula, E V_aj E R_bj and E D_cj q and z represent the field strength vectors at the location of the vehicle j, respectively, representing the kinetic energy field formed by a single moving object, the potential energy field formed by a stationary object, and the behavioral field formed by the driver. p, q, and z are the total number of objects in each field.

[0091] (3) Calculate the force F exerted on the vehicle under test in the driving risk field based on the total field strength of the driving risk field. j for:

[0092] F j =E S_j Mj [R j ·exp(-k2v j cosθ j )·(1+D j (6)

[0093] In the formula, E S_j M represents the total field strength of the driving risk field for the tested vehicle. j R represents the virtual mass of the vehicle being tested. j The road condition factor is the location of the vehicle being tested, k2 is an undetermined coefficient, and v j Let θ be the speed of the vehicle being measured. j The direction of velocity v j With field strength E S_j The angle between directions, D j The risk factors for the tested vehicle.

[0094] (4) Calculate the driving risk coefficient C of the tested vehicle based on the forces acting on the vehicle in the driving risk field. risk for:

[0095]

[0096] In the formula, F j This represents the risk field force exerted on the tested vehicle j at a certain moment within the driving risk field formed by the surrounding environment's elements of people, vehicles, and roads. F and σ F The force F represents the force applied over a certain period of time. j Let the mean and standard deviation be... It conforms to a standard normal distribution, and f(x) represents the density function of the standard normal distribution. Indicates that the independent variable is greater than Probability.

[0097] In practical applications, the kinetic energy field represents the "physical field" that characterizes the impact of moving objects on driving risks. Moving objects on the road mainly include moving vehicles, pedestrians, animals, and non-motorized vehicles. The magnitude and direction of the kinetic energy field are mainly determined by the object's properties, state of motion, and road conditions. Influencing factors include object type, mass, speed, acceleration, road surface adhesion coefficient, and road gradient, primarily reflecting the magnitude of the object's kinetic energy. The potential energy field represents the "physical field" that characterizes the impact of stationary objects on driving risks. Stationary objects on the road mainly include stopped vehicles, medians, roadblocks, and traffic signs. The magnitude and direction of the potential energy field are mainly determined by the properties of the stationary objects and road conditions. Influencing factors include object type, mass, and environmental visibility. The behavioral field represents the "physical field" that characterizes the impact of driver behavior characteristics on driving risks. Driver behavior characteristics mainly include the driver's driving style, skills, legal awareness, gender, age, driving experience, personality, physical fitness, and psychological state. The magnitude and direction of the behavioral field are mainly determined by the driver's behavioral characteristics. For example, aggressive drivers often have a higher risk factor than conservative drivers, and their "behavioral field" intensity is higher. Drivers with low driving skills usually have a higher "behavioral field" than drivers with high driving skills, and so on.

[0098] Step 400: Optimize the driving risk estimation model to obtain an optimized risk estimation model for the test.

[0099] Specifically, the process includes the following steps:

[0100] (1) Based on simulation software, dangerous accident scenarios were reconstructed and tested to obtain verification results.

[0101] In this embodiment, a virtual scene of a dangerous accident is reconstructed in simulation software, and high-risk scenario testing and verification of autonomous driving is carried out in the virtual scene.

[0102] (2) Further analyze driving patterns based on the verification results.

[0103] (3) Adjust the weights in the driving risk estimation model according to driving patterns to optimize the driving risk estimation model.

[0104] Specifically, based on the factors affecting driving risks in the kinetic energy field, potential energy field and behavioral field, the importance judgment is increased (the richness of the field strength type of the "physical field", i.e. the number of influencing factors), and the weight is adjusted, that is, each item in formula (1) is increased with a weight coefficient, so that the final calculated driving risk coefficient is more accurate.

[0105] Step 500: Based on the temporal evolution characteristics of hazardous events, construct a hazardous time chain model according to the optimized risk estimation model.

[0106] include:

[0107] (1) The time series of road traffic accidents and conflicts in the optimized risk estimation model are treated as Markov random processes.

[0108] (2) Based on Markov stochastic processes, a dangerous event chain model is formed by determining the initial probability of different states in a road traffic accident and the transition probability distribution between states in the chain.

[0109] Specifically, Markov stochastic processes utilize Hidden Markov Models (HMMs), which are statistical models used to describe a Markov process containing hidden unknown parameters. A standard HMM has five variables (q, O, X, A, h). Here, q represents the hidden state, indicating a state that cannot be directly observed. O represents the observable state, indicating a state that can be directly observed. In the model, these states have specific relationships with the hidden states. X is the initial state probability matrix, representing the state transition probability distribution of the hidden states at the initial time step. A is the hidden state transition probability matrix, used to describe the probability of transitions between states in the model, where N represents the number of hidden states. ij =P(q) j |q i ), 1≤i,j≥N, indicates that at time t, the state is q. i Under the given conditions, the state at time t+1 is q j The probability of h is the observation state transition probability matrix. Let M represent the number of observable states, and h is the probability of h. ij =P(O i |q j ), 1≤i≤M, 1≤j≤N, indicates that at time t, the hidden state is q. j Under the condition that the observed state is O i The probability of.

[0110] Step 600: Solve the hazard time chain model to obtain the hazard event chain.

[0111] include:

[0112] (1) Based on the road hazard event status and vehicle driving risk coefficient distribution in the hazard event chain model, the range of the road segment passed through at different times is set to represent the hidden state, and the level is divided accordingly.

[0113] (2) Establish a prediction model based on the training dataset, which includes existing road hazard event statuses.

[0114] Specifically, existing road hazard event states are selected as the training dataset to build a prediction model, and future road hazard event states are selected as the test set to verify the accuracy and predictive power of the prediction model, ultimately obtaining a complete hazard event chain.

[0115] (3) Calculate the number of training datasets that change from the state of the previous time step to the state of the next time step, and use the formula to calculate the state transition matrix A and the observation variable transition probability matrix h.

[0116] (4) Determine the initial probability vector based on the state transition matrix A, the observation variable transition probability matrix h, and the prediction model.

[0117] (5) Determine the predicted value of the hidden state based on the prediction model.

[0118] (6) Predict the actual dangerous event state based on the initial probability vector and the hidden state prediction value. The actual dangerous event state is used to construct the dangerous event chain model.

[0119] For the initial probability value, assuming the hidden state at time t-1 is known to predict the state at time t, let {q} t-1 ,q t ,q t+1 The initial probability vector X is represented as {1,0,0}, which means that the range of dangerous event states under the actual complex traffic conditions at time t-1 is assumed to be the value of X. This value is substituted into the prediction model to continuously calculate the latest X value. The hidden state with the highest probability at time t is calculated using a formula, which is the predicted value of the hidden state at time t. Finally, the most realistic dangerous event state for the next time period t+1 is predicted, thus completing the reproduction and reconstruction of traffic accident scenarios and predicting dangerous events in the next time period when event conflicts occur in natural experiments, forming a closed loop for scenario generation.

[0120] In this embodiment, based on the existing types of traffic risk sources, the strong coupling characteristics between different levels and types of risk sources are explored, and the differences in the degree of influence of different risk sources on traffic accidents are analyzed to obtain the risk level and risk weight of each factor. In this way, comprehensive risk elements associated with multiple types of risk sources are extracted, and traffic risks corresponding to specific traffic scenarios are modeled.

[0121] Basic hazard chain model such as Figure 2As shown, based on a large amount of test scenario data, the influence mechanism of different elements on driving risk is explored by analyzing the characteristics of road traffic participants and environmental factors in the human-vehicle-road system. The uncertainty in the test scenario is quantified based on natural driving data, actual sensor performance and vehicle model, including the uncertainty of perception results, the uncertainty of interaction characteristics and the uncertainty of information blind spots. On this basis, the driving risk quantification model considering uncertain information is improved. Based on the virtual scene reconstruction and automated simulation calculation methods used in the accelerated test, the driving rules of multi-source heterogeneous data are fully explored. Based on the natural driving dataset and the driving risk influencing factors in the kinetic energy field, potential energy field and behavioral field, the importance judgment is added and the weight is adjusted. That is, the weight coefficient is added to each term in equation (1), and then the model is lightweighted.

[0122] After optimizing the risk field model, the temporal evolution characteristics of hazardous events are studied, and a hazardous event chain model is constructed. The time series of road traffic accidents / conflicts are treated as Markov random processes. By exploring the initial probabilities of different states of traffic events and the transition probability distributions between states in the chain, the evolution process of the traffic accident chain is described mathematically.

[0123] Furthermore, this embodiment also provides a hazard event chain extraction system based on complex traffic scenarios, including:

[0124] The acquisition module is used to acquire test scenario data;

[0125] The quantification module is used to quantify the uncertainty of test scenario data based on preset natural driving data and sensor performance information, in order to determine the key factors of dynamic attributes and interactions between the vehicle and the environment.

[0126] The estimation model building module is used to build a driving risk estimation model for test scenarios based on a preset multi-dimensional feature set of risk estimation and uncertainty information.

[0127] The first optimization module is used to optimize the driving risk estimation model to obtain an optimized risk estimation model for testing.

[0128] The second optimization module is used to construct a hazard time chain model based on the temporal evolution characteristics of hazardous events and the optimized risk estimation model.

[0129] The parsing module is used to solve the hazard time chain model to obtain the hazard event chain.

[0130] This invention utilizes collected traffic accident data and traffic conflict data collected in natural driving experiments to reveal the spatiotemporal evolution of complex traffic scenarios, studies the method of extracting dangerous event chains, and improves the effect of reproducing and reconstructing test scenarios in the later stage by solving the dangerous time chain model.

[0131] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0132] The above embodiments are only for illustrating the technical concept and features of the present invention. Their purpose is to enable those skilled in the art to understand the content of the present invention and implement it. They should not be used to limit the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for extracting hazardous event chains based on complex traffic scenarios, characterized in that: include: S1, Obtain test scenario data; S2, based on preset natural driving data and sensor performance information, determines the key factors of the dynamic attributes and interactions between the vehicle and the environment, and quantifies the uncertainty of the key factors in a hierarchical manner, including: (1) Obtain the uncertainty of the perception result based on the sensor performance information, which includes camera data, millimeter-wave radar data and lidar data; (2) Based on the natural driving data, obtain the uncertainty of interaction characteristics and the uncertainty of information blind spots; The uncertainty of the perception results refers to the results of individual and fused perception and understanding from cameras, millimeter-wave radar, and lidar. The uncertainty of the interaction characteristics includes at least the randomness of traffic participants in mixed traffic. The uncertainty of the information blind spot includes at least the uncertainty of road conditions, obstruction and absence of traffic facilities. S3, based on a pre-defined multi-dimensional feature set of risk estimation and uncertainty information, constructs a driving risk estimation model for test scenarios, including: (1) Construct a multidimensional feature set for risk estimation based on the key factors of the dynamic attributes and interactions between the vehicle and the environment determined in step S2; (2) Based on the combined effect of each element in the multidimensional feature set of the risk estimation on driving risk, the driving risk estimation model is established. The driving risk estimation model includes the field strength of each traffic element, the force exerted by the risk field on the vehicle under test, and the driving risk coefficient. S4, optimize the driving risk estimation model to obtain an optimized risk estimation model for the test; S5, Based on the temporal evolution characteristics of hazardous events, a hazardous time chain model is constructed according to the optimized risk estimation model, including: (1) The time series of road traffic accidents and conflicts in the optimized risk estimation model are treated as Markov random processes; (2) Based on Markov stochastic processes, a dangerous event chain model is formed by determining the initial probabilities of different states in the road traffic accident and the transition probability distribution between states in the chain; S6, Solve the dangerous time chain model to obtain the dangerous event chain, including: (1) Based on the road hazard event status and vehicle driving risk coefficient distribution in the hazard event chain model, the range of road segments passed through at different times is set to represent the hidden state, and the levels are divided accordingly; (2) Establish a prediction model based on the training dataset, which includes existing road hazard event states; (3) Calculate the number of training datasets that change from the state of the previous time step to the state of the next time step, and calculate the state transition matrix and the observation variable transition probability matrix; (4) Determine the initial probability vector based on the state transition matrix, the observed variable transition probability matrix, and the prediction model; (5) Determine the predicted value of the hidden state based on the prediction model; (6) Predict the actual dangerous event state based on the initial probability vector and the hidden state prediction value.

2. The method for extracting hazard event chains based on complex traffic scenarios according to claim 1, characterized in that: Step S1 includes: (1) Collect traffic accident data, which includes vehicle, road and environmental data of traffic accidents that occur during the natural driving process of a car; (2) Traffic conflict data are collected based on natural driving experiments. The traffic conflict data includes vehicle-to-vehicle conflict data and vehicle-to-road conflict data.

3. The method for extracting hazard event chains based on complex traffic scenarios according to claim 1, characterized in that: Step S4 includes: (1) Based on simulation software, dangerous accident scenarios were reconstructed and tested to obtain verification results; (2) Analyze driving patterns based on the verification results; (3) Adjust the weights in the driving risk estimation model according to the driving rules to optimize the driving risk estimation model.

4. The method for extracting hazard event chains based on complex traffic scenarios according to claim 3, characterized in that: Based on the factors influencing driving risks in the kinetic energy field, potential energy field, and behavioral field, an importance judgment is added, and the weights in the driving risk estimation model are adjusted.

5. A system for extracting hazardous event chains based on complex traffic scenarios, characterized in that: Based on the method of any one of claims 1 to 4, the system comprises: The acquisition module is used to acquire test scenario data; The quantization module is used to determine the key factors of the dynamic attributes and interactions between the vehicle and the environment based on preset natural driving data and sensor performance information, and to perform hierarchical quantification of the uncertainty of the key factors. The estimation model construction module is used to construct a driving risk estimation model for the test scenario based on a preset multidimensional feature set of risk estimation and the uncertainty information. The first optimization module is used to optimize the driving risk estimation model to obtain an optimized risk estimation model for the test. The second optimization module is used to construct a hazard time chain model based on the temporal evolution characteristics of hazardous events and the optimized risk estimation model. The parsing module is used to solve the dangerous time chain model to obtain the dangerous event chain.

Images