Traffic conflict risk quantification method based on probability trajectory prediction and physical modeling

Abstract

The invention relates to a traffic conflict risk quantification method based on probability trajectory prediction and physical modeling. The method comprises the steps of obtaining video data of traffic participants at a target traffic scene and extracting state features to construct a trajectory data set; constructing a space interaction graph sequence, inputting the space interaction graph sequence into a trajectory prediction model of a fusion graph attention network and a Transform network, and generating a plurality of candidate trajectories and occurrence probabilities of the candidate trajectories; traversing candidate trajectory combinations of different traffic participants, identifying potential conflict points by adopting a high-order polynomial fitting trajectory and a numerical solution method, and calculating a conflict arrival time difference to perform conflict risk judgment; and constructing a collision dynamics model to calculate the expected impact energy of collision, and quantifying the risk severity of potential conflicts in combination with the occurrence probability of the predicted trajectory. According to the technical scheme, the uncertainty of future behaviors of traffic participants can be accurately described, and high-precision multi-modal trajectory prediction and comprehensive conflict risk quantitative evaluation are realized.

Description

Technical Field

[0001] This invention relates to the fields of computer technology, autonomous driving trajectory prediction, and road traffic safety technology, specifically to a method for quantifying traffic conflict risk based on probabilistic trajectory prediction and physical modeling. Background Technology

[0002] With the accelerated pace of urbanization, urban traffic flow is showing a continuous growth trend, and various traffic problems are becoming increasingly prominent. Existing traffic conflict detection methods usually assume that traffic participants move along a single deterministic trajectory, ignoring the diversity and uncertainty of traffic participants' intentions in real traffic scenarios. They also rely heavily on rule-based time indicators for conflict determination or only provide binary judgment results, lacking quantification of conflict severity, resulting in an incomplete and inaccurate assessment of conflict risk.

[0003] Currently, many studies use deep learning methods to predict multiple possible future trajectories and assign probabilities to them, enabling a more realistic portrayal of the diverse intentions of traffic participants. However, many methods fail to fully consider the spatiotemporal interactions among traffic participants, and they tend to focus on evaluating the accuracy of trajectory prediction, paying less attention to jointly modeling the prediction results with conflict detection and risk assessment. Therefore, there is an urgent need for a traffic conflict risk quantification method that integrates multimodal trajectory prediction results with physical collision modeling to comprehensively address the problems existing in current methods. Summary of the Invention

[0004] The purpose of this invention is to provide a traffic conflict risk quantification method based on probabilistic trajectory prediction and physical modeling, which can perform multimodal probability prediction of the future behavior of traffic participants and quantify the possibility and severity of potential conflicts, thereby assessing conflict risks more scientifically.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] A method for quantifying traffic conflict risk based on probabilistic trajectory prediction and physical modeling includes the following steps:

[0007] R1: Acquire video data of traffic participants in the target traffic scene and preprocess it to extract the type, position, speed, acceleration and heading angle of all traffic participants to construct a trajectory dataset;

[0008] R2: Construct a spatial interaction graph sequence using trajectory data, and input it into a trajectory prediction model that integrates a graph attention network and a Transformer to predict multiple candidate trajectories for each traffic participant in the future time period and their corresponding probability of occurrence.

[0009] R3: Collision detection is performed based on the uncertainty of the predicted trajectory. It iterates through candidate trajectory combinations from different traffic participants, identifies potential conflict points through high-order polynomial fitting and numerical solutions, and calculates the collision arrival time difference. Conduct conflict risk assessment;

[0010] R4: For trajectory pairs with a risk of collision, construct a collision dynamics model to calculate the expected impact energy of the collision. By combining the predicted trajectory's probability of occurrence, the severity of potential conflict risks can be quantified. .

[0011] Specifically, step R2 is as follows:

[0012] R2-1: For each moment of a continuous time step Constructing a spatial interaction map using trajectory data ,in For a set of nodes, It is an edge set; each traffic participant corresponds to a node. Its eigenvectors Defined as: ,in For position vectors, For velocity vector, For acceleration vectors, For heading angle, Encode the participant type when the node and nodes Distance between When the radius of influence is smaller than the preset space, create a space from... point to Directed edge This indicates that there is a spatial interaction relationship between them;

[0013] R2-2: Spatial interaction graphs at each time step The input is fed into a graph attention network, where the attention weights between nodes are calculated using a multi-head attention mechanism. This weight Represents a node For nodes The degree of influence is determined by aggregating information from neighboring nodes and outputting the spatial dimension of interaction features.

[0014] R2-3: Spatial interaction features are aggregated into a feature sequence through pooling operations and then input into a Transformer encoder. The Transformer captures long-term temporal dependencies in the sequence through a self-attention mechanism. After decoding, the sequence is mapped through a fully connected layer to generate the future time period. Candidate trajectories and their corresponding occurrence probabilities Each candidate trajectory is for the future Location sequence at each time step The probability of the candidate trajectory satisfies .

[0015] Specifically, step R3 is as follows:

[0016] R3-1: Iterate through any two traffic participants and Trajectory mode combination , and They are respectively and The The first mode and the first The trajectory of each mode is used to fit the sequence of candidate trajectory points into a parametric curve using a high-order polynomial. Preferably, a fifth-order polynomial is used. Perform fitting;

[0017] R3-2: Solving the trajectory modal combination using numerical methods Fitting polynomial equations Preferably, the solution to the equation is found by using Newton's iterative method to continuously approximate the roots of the function: A convergence criterion is set and constrained by the coordinate range of the trajectory to obtain the coordinates of the geometric intersection points of the trajectory in the prediction time domain. These are potential points of conflict.

[0018] R3-3: The fitting polynomial for the trajectory and Points-based calculation of traffic participants and The actual path length from the current location to the point of conflict is used to calculate the arrival time, combined with the current speed, and the time difference between the two collisions is used as the basis for calculation. As a criterion for determining conflict:

[0019] ;

[0020] in, and They are respectively and At the x-coordinate position of the prediction start time, and The first derivative of the fitted curve. and They are respectively and The current speed; when When the value is less than a preset safety threshold, the trajectory mode combination is determined to be... There is a risk of conflict among traffic participants.

[0021] Specifically, step R4 is as follows:

[0022] R4-1: Combination of trajectory modes with conflict risk Based on the inelastic collision assumption, considering emergency braking and avoidance behaviors during a collision, the participants are calculated. and velocity change and :

[0023] ;

[0024] ;

[0025] in, and They are respectively and quality and The magnitude of the velocity at the moment of collision. The angle between the velocity directions, For emergency braking and avoidance time, and Each is its maximum deceleration;

[0026] R4-2: Based on the aforementioned velocity change and Calculate the expected impact energy of the collision. : Combined with this conflict trajectory mode combination The corresponding trajectory probability is used to calculate the severity of the collision risk. : ,in, and Participants and The The first mode and the first The probability of trajectory occurrence for each modality, based on the risk severity value. The risk of conflict is classified into different levels.

[0027] The traffic conflict risk quantification method based on probabilistic trajectory prediction and physical modeling provided in the above technical solution has the following advantages compared with existing technologies:

[0028] 1. This invention utilizes a trajectory prediction model that fuses graph attention networks and Transformer networks. This model can simultaneously capture the complex spatial interactions and temporal dependencies among traffic participants, thereby generating multimodal probabilistic trajectories that better reflect actual driving decisions. This improves trajectory prediction accuracy while accurately characterizing the uncertainty and multimodal features of future motion. Furthermore, by traversing all possible trajectory modal combinations for conflict detection, it overcomes the limitations of single deterministic prediction and more comprehensively identifies potential conflicts.

[0029] 2. This invention uses high-order polynomial fitting and numerical methods to accurately locate conflict points, improving the reliability of conflict determination. It also combines the trajectory uncertainty obtained from the trajectory prediction model with the physics-based collision dynamics model to achieve a refined and quantitative analysis of the severity of conflict risks, providing a more reliable basis for risk assessment. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating the traffic conflict risk quantification method based on probabilistic trajectory prediction and physical modeling of the present invention.

[0031] Figure 2 This is a schematic diagram of the construction of a spatial interaction graph in an embodiment of the present invention;

[0032] Figure 3 This is a schematic diagram of trajectory fitting and conflict point identification in an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0034] This embodiment uses the scene at the intersection of Haitang Bay in Hefei City as an example, and combines it with the attached... Figures 1-3 The present invention will be further analyzed and explained.

[0035] The traffic conflict risk quantification method based on probabilistic trajectory prediction and physical modeling includes the following steps:

[0036] Step 1: Acquire video data of traffic participants at the intersection and preprocess it to extract the type, location, speed, acceleration and heading angle of all traffic participants to construct a trajectory dataset;

[0037] Step 2: Construct a spatial interaction graph sequence using the trajectory data obtained in Step 1, and input it into the trajectory prediction model that integrates graph attention network and Transformer to predict multiple candidate trajectories for each traffic participant in the future time period and their corresponding probability of occurrence.

[0038] 2.1 For each moment of a continuous time step Constructing a spatial interaction map using trajectory data ,in For a set of nodes, It is an edge set; each traffic participant corresponds to a node. Its eigenvectors Information including position, velocity, acceleration, heading angle, and type encoding, when the distance between two nodes... When the spatial influence radius is smaller than the preset value, create a directed edge. Representing spatial interaction relationships, such as Figure 2 As shown;

[0039] 2.2 Spatial interaction graph of the past 15 time steps The input is fed into a graph attention network, where the attention weights between nodes are calculated using a multi-head attention mechanism. It aggregates information from neighboring nodes and outputs the spatial dimension of interaction features;

[0040] 2.3 Spatial interaction features are aggregated into a feature sequence through pooling operations and then input into the Transformer encoder to capture long-term temporal dependencies in the sequence. After decoding, the sequence is mapped through a fully connected layer to generate the features for the next 25 time steps. Candidate trajectories and their corresponding occurrence probabilities.

[0041] Step 3: Detect collisions based on the uncertainty of the predicted trajectory. Traverse candidate trajectory combinations from different traffic participants, identify potential conflict points using high-order polynomial fitting and numerical solutions, and calculate the collision arrival time difference. Conduct conflict risk assessment;

[0042] 3.1 For any two traffic participants, traversing their respective 5 candidate trajectories, there are a total of For each combination of trajectory modes, The trajectory in the equation uses a fifth-order polynomial. Perform fitting;

[0043] 3.2 Solving the trajectory mode combination using Newton's iterative method Fitting polynomial equations The potential conflict points of the predicted trajectory in the time domain are obtained. ;

[0044] 3.3 Polynomial for Trajectory Fitting and Points-based calculation of traffic participants and The actual path length from the current location to the point of conflict is used to calculate the arrival time, combined with the current speed, and the time difference between the two collisions is used as the basis for calculation. As a criterion for determining conflict:

[0045] ;

[0046] In this embodiment, the safety threshold is set to 2.0 seconds. If the value is less than 2.0 seconds, the trajectory mode combination is determined. There is a risk of conflict among traffic participants.

[0047] Step 4: For trajectory pairs with a risk of collision, construct a collision dynamics model to calculate the expected impact energy of the collision. By combining the predicted trajectory's probability of occurrence, the severity of potential conflict risks can be quantified. ;

[0048] 4.1 For trajectory mode combinations deemed to have a risk of conflict Based on the inelastic collision assumption, and considering traffic participants... The participants had a reaction time of seconds to take emergency braking and avoid danger, and the calculation of the participants' actions was performed. and velocity change and :

[0049] ;

[0050] ;

[0051] 4.2 Based on the aforementioned velocity change and Calculate the expected impact energy of the collision. : Combined with this conflict trajectory mode combination The corresponding trajectory probability is used to calculate the severity of the collision risk. : ,in, and Participants and The The first mode and the first The probability of trajectory occurrence for each modality. Based on risk severity values. The risk of conflict is classified into different levels.

[0052] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. For those skilled in the art, after learning the contents described in the present invention, several equivalent changes and substitutions can be made without departing from the principle of the present invention. These equivalent changes and substitutions should also be considered to fall within the protection scope of the present invention.

Claims

1. A method for quantifying traffic conflict risk based on probabilistic trajectory prediction and physical modeling, characterized in that, Includes the following steps: R1: Acquire video data of traffic participants in the target traffic scene and preprocess it to extract the type, position, speed, acceleration and heading angle of all traffic participants to construct a trajectory dataset; R2: Construct a spatial interaction graph sequence using trajectory data, and input it into a trajectory prediction model that integrates a graph attention network and a Transformer to predict multiple candidate trajectories for each traffic participant in the future time period and their corresponding probability of occurrence. R3: Collision detection is performed based on the uncertainty of the predicted trajectory. It iterates through candidate trajectory combinations from different traffic participants, identifies potential conflict points through high-order polynomial fitting and numerical solutions, and calculates the collision arrival time difference. Conduct conflict risk assessment; R4: For trajectory pairs with a risk of collision, construct a collision dynamics model to calculate the expected impact energy of the collision. By combining the predicted trajectory's probability of occurrence, the severity of potential conflict risks can be quantified. .

2. The traffic conflict risk quantification method based on probabilistic trajectory prediction and physical modeling according to claim 1, characterized in that, Step R2 is as follows: R2-1: For each moment of a continuous time step Constructing a spatial interaction map using trajectory data ,in For a set of nodes, It is an edge set; each traffic participant corresponds to a node. , Its eigenvectors Defined as: ,in For position vectors, For velocity vectors, For acceleration vectors, For heading angle, Encode the participant type when the node and nodes The distance between them satisfies When creating from point to directed edges , Radius of spatial influence; R2-2: Spatial interaction graphs at each time step The input is fed into a graph attention network, where the attention weights between nodes are calculated using a multi-head attention mechanism. This weight Represents a node For nodes The degree of influence is determined by aggregating information from neighboring nodes and outputting the spatial dimension of interaction features. R2-3: Spatial interaction features are aggregated into a feature sequence through pooling operations and then input into a Transformer encoder. The Transformer captures long-term temporal dependencies in the sequence through a self-attention mechanism. After decoding, the sequence is mapped through a fully connected layer to generate the future time period. Candidate trajectories and their corresponding occurrence probabilities .

3. The traffic conflict risk quantification method based on probabilistic trajectory prediction and physical modeling according to claim 1, characterized in that, Step R3 is as follows: R3-1: Iterate through any two traffic participants and Trajectory mode combination , and They are respectively and The The first mode and the first The trajectory of each mode is used to fit the sequence of candidate trajectory points into a parametric curve using a high-order polynomial. ; R3-2: Solving the trajectory modal combination using numerical methods Fitting polynomial equations Obtain the coordinates of the geometric intersection of the trajectories within the predicted time domain. These are potential points of conflict. R3-3: The fitting polynomial for the trajectory and Points-based calculation of traffic participants and The actual path length from the current location to the point of conflict is used to calculate the arrival time, combined with the current speed, and the time difference between the two collisions is used as the basis for calculation. As a criterion for determining conflict: ； in, and They are respectively and The x-coordinate position at the prediction start time, and The first derivative of the fitted curve. and They are respectively and The current speed; when When the value is less than a preset safety threshold, the trajectory mode combination is determined to be... There is a risk of conflict among traffic participants.

4. The traffic conflict risk quantification method based on probabilistic trajectory prediction and physical modeling according to claim 1, characterized in that, Step R4 is as follows: R4-1: Combination of trajectory modes with conflict risk Based on the inelastic collision assumption, the participants are calculated. and velocity change and : ； ； in, and They are respectively and quality and The magnitude of the velocity at the moment of collision. The angle between the velocity directions, For emergency braking and avoidance time, and Each is its maximum deceleration; R4-2: Based on the aforementioned velocity change and Calculate the expected impact energy of the collision. : Combined with this conflict trajectory mode combination The corresponding trajectory probability is used to calculate the severity of the collision risk. : ,in, and Participants and The The first mode and the first The probability of trajectory occurrence for each modality, based on the risk severity value. The risk of conflict is classified into different levels.

Images