Dynamic weight correction of vehicle trajectory and path deviation probability prediction method

Abstract

The present invention discloses a method for dynamic weight correction of vehicle trajectory and prediction of path deviation probability. The method collects vehicle data and multi-source dynamic data of vehicle trajectory in real time through optical sensors, radio wave sensors and inertial navigation sensors, performs spatiotemporal calibration, extracts obstacle features and road structure features, quickly captures obstacle movement trends through spatiotemporal graph sequences and graph convolutional neural networks, realizes real-time obstacle avoidance in combination with dynamic weight adjustment, predicts driving intentions using Bayesian neural networks, adjusts trajectory planning strategies through weight correction, reduces uncertainty through multi-source data fusion and probabilistic prediction, and continuously optimizes the model through a closed-loop feedback mechanism. Through real-time weight adjustment and closed-loop feedback, the method maintains efficient operation in complex scenarios such as intersections and roundabouts, thereby achieving the purpose of being able to quickly respond to dynamic obstacles or changes in driving behavior and improving the accuracy of path deviation prediction probability.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle trajectory prediction, and in particular relates to a method for dynamic weight correction of vehicle trajectories and path deviation probability prediction. Background Art

[0002] In autonomous driving systems, dynamic correction of vehicle trajectories and prediction of path deviation probabilities are particularly important. By predicting deviation probabilities in real time, the autonomous driving system can trigger emergency avoidance or deceleration strategies in advance. When it is predicted that the vehicle may deviate from the drivable area, the trajectory is regenerated by adjusting the path planning weights to ensure that the vehicle is always within the safety boundary and reduce the risk of collision.

[0003] Currently, dynamic correction of autonomous driving paths is typically performed through path planning correction based on multi-source data fusion. For example, patent application number CN119063749A discloses a method, apparatus, and device for dynamic correction of autonomous driving path planning. The method determines the current vehicle's drivable area by acquiring point cloud data from a millimeter-wave radar and image data from a visual camera. A grid map with composite weights is constructed, and a global reference path is obtained by searching the grid map using the Dijkstra algorithm. A dynamic programming method is used to determine the current vehicle's distance from the global reference path. A spline curve is fitted to the sampling point set. The drivable area is convexified using a scattered sampling method based on the spline curve to obtain a convex spatial boundary of the drivable area. A quadratic programming problem is constructed based on the convex spatial boundary to solve the desired trajectory. When the reference trajectory is detected to deviate from the drivable area, the weights of the sampling point set and the quadratic programming problem are adjusted to correct the desired trajectory, thereby improving the accuracy of the transmitted desired trajectory.

[0004] However, the dynamic programming method used to determine the set of sampling points of the current vehicle's distance from the global reference path is based on a grid map searched through the Dijkstra algorithm to obtain the global reference path. Grid maps usually rely on real-time scanning and updating by sensors (such as millimeter-wave radar). However, the positions of dynamic obstacles (such as pedestrians and moving objects) change frequently, resulting in the map update speed being unable to keep up with environmental changes, making it poorly adaptable to dynamic environments. Therefore, we need to propose a dynamic weight correction method for vehicle trajectories and a path deviation probability prediction method to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for dynamic weight correction and path deviation probability prediction of vehicle trajectories, which can quickly respond to dynamic obstacles or changes in driving behavior, improve the accuracy of path deviation prediction probability, and thus improve adaptability to dynamic environments, so as to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The method for dynamic weight correction of vehicle trajectory and path deviation probability prediction includes the following steps:

[0008] S1. Use optical sensors, radio wave sensors, and inertial navigation sensors to collect real-time vehicle data and multi-source dynamic data of vehicle trajectories, perform spatiotemporal calibration, and extract obstacle features and road structure characteristics.

[0009] S2, construct a spatiotemporal graph sequence with obstacles as nodes and spatial distance and time correlation as edges;

[0010] S3, combining obstacle features and road structure features to predict the ideal trajectory of the vehicle without obstacle interference;

[0011] S4, predicting the future trajectory of obstacles through graph convolution and spatiotemporal attention mechanism, and integrating it with the ideal trajectory and road structure features to form an interactive trajectory set;

[0012] S5. Define a path deviation probability threshold based on the road structure or navigation path, and use a Bayesian neural network to predict the path deviation probability and uncertainty interval of future frames;

[0013] S6. defining a state space according to the prediction results of the Bayesian neural network, and modifying the data fusion weight according to the defined state space;

[0014] S7. Monitor the corrected trajectory in real time and calculate the deviation from the actual trajectory. Iteratively modify the weight strategy based on the deviation to form a closed-loop feedback loop of prediction-evaluation-correction.

[0015] Preferably, the optical sensor uses a high-definition camera to collect image data and identifies the type and position of obstacles through a target detection algorithm; the radio wave sensor uses a millimeter-wave radar to obtain the distance, speed and reflection intensity of obstacles; and the inertial navigation sensor obtains the vehicle's own motion reference through an inertial measurement unit and a global navigation satellite system as a reference coordinate system for trajectory prediction.

[0016] Preferably, when performing spatiotemporal calibration, the timestamps of different sensors are aligned through the hardware clock, and then the data of each sensor are unified into the world coordinate system, and the coordinate transformation is completed through the extrinsic parameter matrix.

[0017] Preferably, when extracting obstacle features, the camera image and millimeter-wave radar point cloud are combined, and the two-dimensional bounding box and the three-dimensional obstacle cluster are associated through spatiotemporal calibration. The obstacle category and position collected by the optical sensor are detected using the target detection model. The obstacle distance, speed and reflection intensity collected by the millimeter-wave radar are extracted through the clustering algorithm. Finally, the obstacle position, size, speed, category and spatiotemporal context information are output;

[0018] When extracting road structure features, the vehicle posture is matched with the high-precision map, the static elements of lane lines and traffic signs are extracted, the image is segmented using a semantic segmentation model, the drivable area and lane lines are identified, and the road structure features such as lane line equations and drivable area masks are output by combining map and visual information.

[0019] Preferably, the process of constructing the spatiotemporal graph sequence is as follows:

[0020] A1. Define each obstacle as a node and connect nodes with spatiotemporal correlation.

[0021] A2. Calculate spatial distance weight, time interval weight and comprehensive edge weight based on the defined nodes;

[0022] A3. Connect the graphs of consecutive time steps along the time axis to form a dynamic graph sequence.

[0023] Preferably, the ideal trajectory is based on the lane centerline or the navigation path, and is combined with the vehicle kinematic model to generate an ideal trajectory without obstacle interference.

[0024] Preferably, the process of predicting the future trajectory of an obstacle is as follows:

[0025] B1. Use graph convolutional networks to process spatiotemporal graphs and capture the interaction features between obstacles;

[0026] B2. Introduce the attention mechanism to focus on key obstacles and calculate the attention weights between nodes;

[0027] B3. Based on the captured interaction features between obstacles and attention weights, the future trajectory is predicted through a recurrent neural network.

[0028] Preferably, the path deviation probability threshold is defined by setting a threshold based on the road structure and vehicle status. If the lateral deviation of the trajectory from the reference path exceeds the set threshold, it is considered a deviation. If the lateral deviation of the trajectory from the reference path is within the set threshold range, it is considered not to have deviated.

[0029] Preferably, when predicting the probability of future frame path deviation, a Bayesian neural network is used to predict the probability of future frame path deviation, and the interaction trajectory set, vehicle state information and environmental characteristics are used as inputs of the Bayesian neural network. The posterior distribution of path deviation is estimated through a probability model, and multiple prediction results are obtained through multiple sampling. The mean and variance of the path deviation probability are calculated, and a probabilistic prediction result is provided through the Bayesian neural network to quantify the uncertainty of the prediction.

[0030] Preferably, the state space includes the deviation probability, uncertainty, and sensor reliability index output by the Bayesian neural network; when correcting the data fusion weight, an adaptive weighted fusion algorithm is used based on the state space, and the weight model is dynamically adjusted in combination with the deviation probability in the state space.

[0031] The method for dynamic weight correction of vehicle trajectories and prediction of path deviation probability proposed in this invention has the following advantages over the prior art:

[0032] 1. The present invention uses spatiotemporal graph sequences and graph convolutional neural networks to quickly capture obstacle movement trends, combines dynamic weight adjustment to achieve real-time obstacle avoidance, uses Bayesian neural networks to predict driving intentions, adjusts trajectory planning strategies through weight correction, reduces uncertainty through multi-source data fusion and probabilistic prediction, and continuously optimizes the model through a closed-loop feedback mechanism. Through real-time weight adjustment and closed-loop feedback, the system maintains efficient operation in complex scenarios such as intersections and roundabouts, thereby achieving the purpose of being able to quickly respond to dynamic obstacles or changes in driving behavior, improving the accuracy of path deviation prediction probabilities, and thus improving adaptability to dynamic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Shows a flowchart according to an embodiment of the present invention;

[0034] Figure 2 A flowchart of a spatiotemporal graph sequence construction process according to an embodiment of the present invention is shown;

[0035] Figure 3 A flowchart of future obstacle trajectory prediction according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0037] The present invention provides Figure 1-3 The method for dynamic weight correction of vehicle trajectory and path deviation probability prediction shown includes the following steps:

[0038] S1. Use optical sensors, radio wave sensors, and inertial navigation sensors to collect real-time vehicle data and multi-source dynamic data of vehicle trajectories, perform spatiotemporal calibration, and extract obstacle features and road structure characteristics.

[0039] Optical sensors use high-definition cameras to collect image data and identify obstacle types and locations through target detection algorithms. Radio wave sensors use millimeter-wave radar to obtain obstacle distance, speed, and reflection intensity. Inertial navigation sensors use inertial measurement units and global navigation satellite systems to obtain the vehicle's own motion reference, which serves as the reference coordinate system for trajectory prediction. They integrate optical sensors (such as cameras), radio wave sensors (such as millimeter-wave radar), and inertial navigation sensors (IMUs) to collect vehicle status (speed, acceleration, attitude) and trajectory data (position, direction) in real time. Multi-source data fusion captures obstacle location, speed, and road structure characteristics (such as lane lines and curvature), providing a basis for dynamic weight adjustment.

[0040] During spatiotemporal calibration, the timestamps of different sensors are aligned through the hardware clock, and then the sensor data are unified into the world coordinate system. The coordinate transformation is completed through the external parameter matrix. The coordinate transformation formula is:

[0041] ,in, is the point in the transformed coordinate system, is the rotation matrix, is the translation vector, For points in the sensor coordinate system, time delays and spatial deviations of multi-sensor data are eliminated through spatiotemporal calibration operations to ensure data consistency. Spatiotemporal calibration avoids prediction errors caused by data misalignment and improves the accuracy of subsequent steps.

[0042] When extracting obstacle features, we combine camera images and millimeter-wave radar point clouds, associate 2D bounding boxes with 3D obstacle clusters through spatiotemporal calibration, and use a YOLOv8 object detection model to detect the obstacle category and location captured by the optical sensor. A clustering algorithm is used to extract the obstacle distance, speed, and reflection intensity captured by the millimeter-wave radar. Finally, the obstacle location, size, speed, category, and spatiotemporal context are output.

[0043] When extracting road structure features, the vehicle posture is matched with the high-precision map, and static elements such as lane lines and traffic signs are extracted. The image is segmented using a semantic segmentation model (DeepLabv3 model) to identify drivable areas and lane lines. The road structure features such as lane line equations and drivable area masks are output by combining map and visual information.

[0044] S2. A spatiotemporal graph sequence is constructed using obstacles as nodes and spatial distance and temporal correlation as edges. For example, in a vehicle-following scenario, the leading vehicle is used as a node, and the relative distance and speed difference between it and the following vehicle serve as edge weights. This spatiotemporal graph sequence effectively captures the movement trends and interaction patterns of obstacles, providing structured support for predicting their future trajectories. Spatial distance and temporal correlation jointly define edge weights, reflecting the real-time impact between obstacles.

[0045] like Figure 2 As shown in Figure 2, the process of constructing the spatiotemporal graph sequence is as follows:

[0046] A1. Define each obstacle as a node and connect nodes with spatiotemporal correlation.

[0047] A2. Calculate the spatial distance weight, time interval weight and comprehensive edge weight based on the defined nodes.

[0048] The calculation formula for spatial distance weight is:

[0049] ,in, is the spatial distance weight value, and are two node positions, is the spatial scale parameter;

[0050] The time interval weight calculation formula is:

[0051] ,in, is the time interval weight value, and are two node positions, is the time scale parameter;

[0052] The formula for calculating the comprehensive edge weight is:

[0053] ,in, is the comprehensive edge weight value;

[0054] A3. Connect the graphs of consecutive time steps along the time axis to form a dynamic graph sequence;

[0055] By modeling the dynamic environment as a structured graph, it is easy to capture the spatiotemporal interaction relationship between obstacles. The graph sequence can represent the temporal dynamics of the environment and provide an interactive modeling basis for subsequent trajectory prediction.

[0056] S3, combining obstacle features and road structure features to predict the ideal trajectory of the vehicle without obstacle interference;

[0057] The ideal trajectory is based on the lane centerline or navigation path and is combined with the vehicle kinematic model to generate an ideal trajectory without obstacle interference. The ideal trajectory prediction formula is:

[0058] ,in, is the velocity component in the x direction, is the velocity component in the y direction, is the vehicle speed, is the vehicle steering angle, is the vehicle wheelbase, is the vehicle heading angle, is the heading angle The rate of change of , that is, the angular velocity;

[0059] By taking the ideal trajectory as a benchmark to provide a reference for subsequent deviation probability prediction and weight correction, the planning-based approach takes road constraints into account to ensure the feasibility of the predicted trajectory.

[0060] S4, predicting the future trajectory of obstacles through graph convolution and spatiotemporal attention mechanism, and integrating it with the ideal trajectory and road structure features to form an interactive trajectory set;

[0061] like Figure 3 As shown in Figure 2, the process of predicting the future trajectory of an obstacle is as follows:

[0062] B1. Use graph convolutional networks to process spatiotemporal graphs and capture the interaction features between obstacles. The formula for capturing the interaction features between obstacles is:

[0063] ,in, For the +1 Layer node features, is the weight matrix, is the activation function, For the Layer node features, is the degree matrix, is the adjacency matrix with self-loops added;

[0064] B2. Introduce the attention mechanism to focus on key obstacles and calculate the attention weights between nodes;

[0065] The attention weight calculation formula is:

[0066] ,in, For nodes For Node The attention weight of the attention level, For nodes The query vector, is the key vector The transpose of is the exponential form of the similarity score, is a normalization term used to normalize all nodes Related nodes Sum the similarity scores to ensure that the sum of the attention weights is 1;

[0067] B3. Based on the captured interaction features between obstacles and attention weights, the future trajectory is predicted through a recurrent neural network. The future trajectory prediction formula is:

[0068] ,in, For obstacles In the future Predicted position at any moment, For the The historical feature vector of an obstacle from time 1 to time t, It is a composite module of graph convolutional network and attention mechanism, used to extract the spatial interaction features and key information weights of obstacles; A recurrent neural network for processing time series feature sequences to generate future trajectory predictions;

[0069] The algorithm formula for integrating the future trajectory of the obstacle with the ideal trajectory and road structure characteristics is:

[0070] ,in, is the interaction trajectory, For the ideal trajectory, is the ideal trajectory weight, For obstacles The attention weight, For obstacles The predicted trajectory, N is the total number of obstacles;

[0071] Graph convolution and attention mechanisms are used to efficiently process spatiotemporal dependencies and improve the accuracy of obstacle trajectory prediction. The interactive trajectory set covers a variety of possible scenarios and provides redundant options for path planning.

[0072] S5. Define a path deviation probability threshold based on the road structure or navigation path, and use a Bayesian neural network to predict the path deviation probability and uncertainty interval of future frames;

[0073] The path deviation probability threshold is defined based on the road structure and vehicle status. If the lateral deviation of the trajectory from the reference path exceeds the set threshold, it is considered a deviation. If the lateral deviation of the trajectory from the reference path is within the set threshold, it is considered as no deviation.

[0074] When predicting the probability of future frame path deviation, a Bayesian neural network (BNN) is used to predict the probability of future frame path deviation. The interaction trajectory set, vehicle state information (such as speed, position, and posture), and environmental characteristics (such as road structure and obstacle distribution) are used as BNN inputs. The posterior distribution of path deviation is estimated through a probabilistic model. BNN is implemented using Monte Carlo dropout (MC Dropout) or variational inference methods. Multiple prediction results are obtained through multiple sampling. The mean and variance of the path deviation probability are calculated. The Bayesian neural network provides probabilistic prediction results, quantifies the uncertainty of the prediction (such as confidence interval), and provides confidence support for decision-making. Probabilistic prediction enables the system to assess risk.

[0075] When predicting uncertainty intervals, the mean and variance of the path deviation probability are calculated based on multiple prediction results of the Bayesian neural network to form an uncertainty interval. The mean represents the estimated value of the path deviation probability, and the variance reflects the degree of uncertainty in the prediction.

[0076] S6. defining a state space according to the prediction results of the Bayesian neural network, and modifying the data fusion weight according to the defined state space;

[0077] The state space includes the deviation probability, uncertainty, and sensor reliability index output by the Bayesian neural network. When correcting the data fusion weight, an adaptive weighted fusion algorithm is used based on the state space to dynamically adjust the weight model in combination with the deviation probability in the state space. The formula for dynamically adjusting the weight model is:

[0078] ,in, To update the weight after combining the trajectory credibility and weight adjustment, is the weight before updating, The learning rate to control the weight update step size, is the path deviation probability, For the The weight adjustment amount of each sensor;

[0079] Dynamically adjust sensor weights based on environmental uncertainty to improve data fusion accuracy. Increase the weight of reliable sensors in high-probability deviation scenarios to reduce the risk of misjudgment.

[0080] S7: Monitor the corrected trajectory in real time and calculate the deviation from the actual trajectory. It iteratively adjusts the weight strategy based on the deviation to form a closed-loop feedback loop of prediction, evaluation, and correction.

[0081] The formula for iterative correction of weight strategy is:

[0082] ,in, For the +1 The weight parameter value after the iteration update, For the The weight parameter value at the iteration, is the learning rate used to control the step size of each weight update, The objective function is about The derivative vector of is the objective function used to measure the error between the model prediction result and the actual value;

[0083] By constructing a spatiotemporal graph sequence with obstacles as nodes and spatiotemporal correlations as edges, the real-time position, speed, and interaction relationship of obstacles can be directly captured, reflecting environmental changes without waiting for grid map updates. The future trajectory of obstacles is predicted through graph convolution and spatiotemporal attention mechanisms, allowing for early perception of obstacle movement trends. Dynamic fusion weight correction allows for real-time adjustment of the fusion ratio of optical sensor, radio wave sensor, and inertial sensor data based on the path deviation probability and uncertainty interval. For example, when the obstacle trajectory prediction shows high dynamics (such as a vehicle cutting into the lane), the weight of radar data can be increased to reduce dependence on the static position of the obstacle in the grid map. If the Bayesian network predicts a high probability of path deviation, the weight of inertial navigation data is increased to strengthen the vehicle's kinematic constraints and reduce dependence on road boundaries in the grid map.

[0084] Through a prediction-evaluation-correction closed loop, the deviation between the corrected trajectory and the actual trajectory is calculated in real time, and the weighting strategy is dynamically adjusted. For example, if the vehicle is found to have deviated from the path because the grid map has not updated the obstacle position in a timely manner, the closed-loop feedback can quickly reduce the weight of the grid map data to avoid error accumulation. Through multiple iterations, the system can learn a more optimal weight distribution strategy in dynamic obstacle scenarios, such as prioritizing obstacle trajectories predicted by radar rather than historical positions in the grid map. This mechanism can adaptively compensate for the impact of grid map update delays. That is, when the environment is highly dynamic, such as in dense pedestrian scenes, the system automatically increases the weight of the prediction model and reduces its reliance on real-time grid updates. When the environment is relatively static, trust in sensor data is restored, balancing computational efficiency and accuracy.

[0085] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for dynamic weight correction of vehicle trajectories and prediction of path deviation probability, characterized by: The steps include: S1. Use optical sensors, radio wave sensors, and inertial navigation sensors to collect real-time vehicle data and multi-source dynamic data of vehicle trajectories, perform spatiotemporal calibration, and extract obstacle features and road structure characteristics. S2, construct a spatiotemporal graph sequence with obstacles as nodes and spatial distance and time correlation as edges; The process of constructing the space-time graph sequence is as follows: A1. Define each obstacle as a node and connect nodes with spatiotemporal correlation. A2. Calculate spatial distance weight, time interval weight and comprehensive edge weight based on the defined nodes; A3. Connect the graphs of consecutive time steps along the time axis to form a dynamic graph sequence; S3, combining obstacle features and road structure features to predict the ideal trajectory of the vehicle without obstacle interference; S4, predicting the future trajectory of obstacles through graph convolution and spatiotemporal attention mechanism, and integrating it with the ideal trajectory and road structure features to form an interactive trajectory set; The process of predicting the future trajectory of an obstacle is as follows: B1. Use graph convolutional networks to process spatiotemporal graphs and capture the interaction features between obstacles; B2. Introduce the attention mechanism to focus on key obstacles and calculate the attention weights between nodes; B3. Based on the captured interaction features between obstacles and attention weights, the future trajectory is predicted through a recurrent neural network; S5. Define a path deviation probability threshold based on the road structure or navigation path, and use a Bayesian neural network to predict the path deviation probability and uncertainty interval of future frames; S6. defining a state space according to the prediction results of the Bayesian neural network, and modifying the data fusion weight according to the defined state space; S7. Monitor the corrected trajectory in real time and calculate the deviation from the actual trajectory. Iteratively modify the weight strategy based on the deviation to form a closed-loop feedback loop of prediction-evaluation-correction.

2. The method for dynamic weight correction and path deviation probability prediction of vehicle trajectory according to claim 1, characterized in that: The optical sensor uses a high-definition camera to collect image data and identifies the type and location of obstacles through a target detection algorithm; The radio wave sensor uses millimeter-wave radar to obtain the distance, speed and reflection intensity of obstacles; the inertial navigation sensor obtains the vehicle's own motion benchmark through the inertial measurement unit and the global navigation satellite system as the reference coordinate system for trajectory prediction.

3. The method for dynamic weight correction and path deviation probability prediction of vehicle trajectory according to claim 2, characterized in that: During spatiotemporal calibration, the timestamps of different sensors are aligned through the hardware clock, and then the sensor data are unified into the world coordinate system, and the coordinate transformation is completed through the external parameter matrix.

4. The method for dynamic weight correction and path deviation probability prediction of vehicle trajectory according to claim 3, characterized in that: When extracting obstacle features, the system combines camera images and millimeter-wave radar point clouds, associates 2D bounding boxes with 3D obstacle clusters through spatiotemporal calibration, uses an object detection model to detect the obstacle category and location captured by the optical sensor, and uses a clustering algorithm to extract the obstacle distance, speed, and reflection intensity captured by the millimeter-wave radar. Finally, it outputs the obstacle location, size, speed, category, and spatiotemporal context information. When extracting road structure features, the vehicle posture is matched with the high-precision map, the static elements of lane lines and traffic signs are extracted, the image is segmented using a semantic segmentation model, the drivable area and lane lines are identified, and the road structure features such as lane line equations and drivable area masks are output by combining map and visual information.

5. The method for dynamic weight correction and path deviation probability prediction of vehicle trajectory according to claim 4, characterized in that: The ideal trajectory is based on the lane centerline or navigation path and is combined with the vehicle kinematic model to generate an ideal trajectory without obstacle interference.

6. The method for dynamic weight correction and path deviation probability prediction of vehicle trajectory according to claim 5, characterized in that: The path deviation probability threshold is defined by setting a threshold based on the road structure and vehicle status. If the lateral deviation of the trajectory from the reference path exceeds the set threshold, it is considered a deviation. If the lateral deviation of the trajectory from the reference path is within the set threshold, it is considered no deviation.

7. The method for dynamic weight correction and path deviation probability prediction of vehicle trajectory according to claim 6, characterized in that: When predicting the probability of future frame path deviation, a Bayesian neural network is used to predict the probability of future frame path deviation. The interaction trajectory set, vehicle status information, and environmental characteristics are used as inputs to the Bayesian neural network. The posterior distribution of path deviation is estimated through a probabilistic model. Multiple prediction results are obtained through multiple sampling. The mean and variance of the path deviation probability are calculated. The Bayesian neural network provides probabilistic prediction results and quantifies the uncertainty of the prediction.

8. The method for dynamic weight correction and path deviation probability prediction of vehicle trajectory according to claim 7, characterized in that: The state space includes the deviation probability, uncertainty, and sensor reliability index output by the Bayesian neural network. When correcting the data fusion weight, an adaptive weighted fusion algorithm is used based on the state space to dynamically adjust the weight model in combination with the deviation probability in the state space.

Images