CN-121999609-A - Traffic conflict risk quantification method based on probability track prediction and physical modeling

CN121999609ACN 121999609 ACN121999609 ACN 121999609ACN-121999609-A

Abstract

The invention relates to a traffic conflict risk quantification method based on probability track prediction and physical modeling, which comprises the steps of obtaining video data of traffic participants at a target traffic scene, extracting state characteristics to construct a track data set, constructing a space interaction graph sequence, inputting the space interaction graph sequence into a track prediction model of a fusion graph attention network and a transition network to generate a plurality of candidate tracks and occurrence probabilities thereof, traversing candidate track combinations of different traffic participants, identifying potential conflict points by adopting a high-order polynomial fitting track and numerical solution method, calculating conflict arrival time difference to judge conflict risks, constructing a collision dynamics model to calculate expected impact energy of collision, and quantifying the risk severity of the potential conflict in combination with the occurrence probability of the predicted tracks. The technical scheme can accurately describe uncertainty of future behaviors of traffic participants, and high-precision multi-mode track prediction and comprehensive conflict risk quantitative evaluation are realized.

Inventors

ZENG WEILIANG
Xie Jiaer
LUO YUHUAN

Assignees

广东工业大学

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (4)

1. The traffic conflict risk quantification method based on probability track prediction and physical modeling is characterized by comprising the following steps of: R1, acquiring video data of traffic participants at a target traffic scene, preprocessing, extracting types, positions, speeds, accelerations and course angles of all the traffic participants, and constructing a track data set; R2, constructing a space interaction graph sequence by utilizing track data, inputting the space interaction graph sequence into a track prediction model of a fusion graph attention network and a transducer, and predicting a plurality of candidate tracks of each traffic participant and the occurrence probability corresponding to the candidate tracks in a future time period; R3, performing conflict detection based on uncertainty of predicted trajectories, traversing candidate trajectory combinations of different traffic participants, identifying potential conflict points by a method of fitting trajectories with a high-order polynomial and solving numerical values, and calculating a collision arrival time difference Performing conflict risk judgment; r4, constructing a collision dynamics model for track pairs with collision risks, and calculating expected collision energy of collision In combination with the occurrence probability of the predicted trajectory, the risk severity of the potential conflict is quantified 。
2. The traffic collision risk quantification method based on probability trajectory prediction and physical modeling according to claim 1, wherein the step R2 is specifically: r2-1 for each time of successive time steps Construction of a spatial interaction map using trajectory data Wherein In the case of a set of nodes, Is an edge set, each traffic participant corresponds to a node Its characteristic vector The definition is as follows: Wherein As a vector of the position of the object, As a velocity vector of the velocity vector, As the acceleration vector, the acceleration vector is calculated, In order to be the heading angle, Coding for participant types, when node Sum node The distance between them is satisfied When creating slave Pointing to Directed edge of (2) , The radius is influenced for the space; R2-2 spatial interaction map of each time step Input into the graph attention network, and calculate the attention weight among nodes through a multi-head attention mechanism The weight is Representing nodes Opposite node Aggregating information of neighbor nodes, and outputting interactive characteristics of space dimension; R2-3, aggregating the space interaction features into feature sequences through pooling operation, inputting the feature sequences into a transducer encoder, capturing long time sequence dependency relations in the sequences through a self-attention mechanism by the transducer, decoding, and mapping through a full connection layer to generate a future period of time Candidate trajectory and probability of occurrence corresponding to same 。
3. The traffic collision risk quantification method based on probability trajectory prediction and physical modeling according to claim 1, wherein the step R3 is specifically: R3-1 traverse any two traffic participants And Trajectory modality combination of (a) , And Respectively is And Is the first of (2) Modality and th Fitting a sequence of candidate trajectory points to a parameterized curve using a higher order polynomial for the trajectories of the individual modalities ; R3-2, solving the track modal combination by adopting a numerical method Is a fitting polynomial equation set Obtaining the geometric intersection point coordinates of the tracks in the prediction time domain Namely, potential conflict points; R3-3 polynomial fitting to trajectory And Score computing traffic participants And The actual path length from the current position to the conflict point is combined with the current speed to calculate the arrival time, and the collision arrival time difference is used As a collision determination index: ; Wherein, the And Respectively is And At the abscissa position of the predicted starting moment, And In order to fit the first derivative of the curve, And Respectively is And When (1) the current speed of When the track mode combination is smaller than a preset safety threshold value, judging the track mode combination There is a risk of collision for the lower traffic participants.
4. The traffic collision risk quantification method based on probability trajectory prediction and physical modeling according to claim 1, wherein the step R4 is specifically: r4-1 track mode combination for conflict risk Calculating participants based on inelastic collision assumptions And Speed variation of (2) And : ; ; Wherein, the And Respectively is And Is used for the quality of the (a), And In order to achieve a speed of the moment of collision, Is the included angle between the speed direction and the direction, For the time of emergency braking and avoiding danger, And Is the respective maximum deceleration; R4-2 based on the speed variation And Calculating expected impact energy of collision : Combining the conflict trace mode combination Corresponding track probability, calculating severity value of collision risk : , wherein, And Respectively are participators And Is the first of (2) Modality and th Probability of occurrence of trajectories of individual modalities based on risk severity values Different grades of classification are carried out on the collision risks.

Description

Traffic conflict risk quantification method based on probability track prediction and physical modeling Technical Field The invention relates to the technical fields of computers, automatic driving track prediction and road traffic safety, in particular to a traffic conflict risk quantification method based on probability track prediction and physical modeling. Background With the accelerated development of the urban process, urban traffic flow presents a continuously growing situation, and various traffic problems are increasingly highlighted. The existing traffic conflict detection method generally assumes that traffic participants move along a single deterministic track, ignores the diversity and uncertainty of the intention of the traffic participants in a real traffic scene, and relies on time indexes based on rules to carry out conflict judgment or only gives binary judgment results, and the lack of quantification of the severity of the conflict results in incomplete and accurate assessment of the risk of the conflict. Currently, many studies predict and give probability to a plurality of possible future trajectories through a deep learning method, and can more truly characterize the intended diversity of traffic participants. However, many methods fail to adequately consider the spatiotemporal interaction relationship between traffic participants and their multi-emphasis on trajectory prediction accuracy assessment, less attention being paid to joint modeling of prediction results with collision detection, risk assessment. Therefore, a traffic collision risk quantification method integrating the multi-mode track prediction result and the physical collision modeling is urgently needed to comprehensively solve the problems existing in the existing method. Disclosure of Invention The invention aims to provide a traffic conflict risk quantification method based on probability track prediction and physical modeling, which can conduct multi-mode probability prediction on future behaviors of traffic participants and conduct quantification analysis on the possibility and severity of potential conflicts, so that the conflict risk is evaluated more scientifically. In order to solve the technical problems, the invention adopts the following technical scheme: a traffic conflict risk quantification method based on probability track prediction and physical modeling comprises the following steps: R1, acquiring video data of traffic participants at a target traffic scene, preprocessing, extracting types, positions, speeds, accelerations and course angles of all the traffic participants, and constructing a track data set; R2, constructing a space interaction graph sequence by utilizing track data, inputting the space interaction graph sequence into a track prediction model of a fusion graph attention network and a transducer, and predicting a plurality of candidate tracks of each traffic participant and the occurrence probability corresponding to the candidate tracks in a future time period; R3, performing conflict detection based on uncertainty of predicted trajectories, traversing candidate trajectory combinations of different traffic participants, identifying potential conflict points by a method of fitting trajectories with a high-order polynomial and solving numerical values, and calculating a collision arrival time difference Performing conflict risk judgment; r4, constructing a collision dynamics model for track pairs with collision risks, and calculating expected collision energy of collision In combination with the occurrence probability of the predicted trajectory, the risk severity of the potential conflict is quantified。 Wherein, the step R2 specifically comprises: r2-1 for each time of successive time steps Construction of a spatial interaction map using trajectory dataWhereinIn the case of a set of nodes,Is an edge set, each traffic participant corresponds to a nodeIts characteristic vectorThe definition is as follows: Wherein As a vector of the position of the object,As a velocity vector of the velocity vector,As the acceleration vector, the acceleration vector is calculated,In order to be the heading angle,Coding for participant types, when nodeSum nodeDistance betweenCreating a slave when the space influence radius is smaller than a preset space influence radiusPointing toDirected edge of (2)Indicating that a spatial interaction relationship exists; R2-2 spatial interaction map of each time step Input into the graph attention network, and calculate the attention weight among nodes through a multi-head attention mechanismThe weight isRepresenting nodesOpposite nodeAggregating information of neighbor nodes, and outputting interactive characteristics of space dimension; R2-3, aggregating the space interaction features into feature sequences through pooling operation, inputting the feature sequences into a transducer encoder, capturing long time sequence dependency relations in the sequences through a se