A two-wheeled vehicle collision protection-oriented intelligent vehicle state path decision method

Abstract

The present application relates to the technical field of intelligent driving, in particular to a kind of intelligent car insurance state path decision method for two-wheeled vehicle collision protection, comprising: S1, left and right vision real-time image is collected, and according to left and right vision real-time image, two-wheeled vehicle trajectory prediction is carried out, and two-wheeled vehicle trajectory prediction result is generated;S2, according to two-wheeled vehicle trajectory prediction result, personnel injury prediction is carried out, and personnel injury prediction result is generated;S3, according to personnel injury prediction result, intelligent car insurance state path decision is made.The present application can make intelligent car insurance state path decision to solve the problem of how to reduce the harm to personnel in the dangerous situation where collision is inevitable.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology, specifically to an intelligent vehicle hazard path decision-making method for collision protection of two-wheeled vehicles. Background Technology

[0002] With the development of autonomous driving technology, intelligent vehicles are able to perceive dangerous traffic scenarios in advance and avoid them, reducing the occurrence of accidents. However, even highly automated driving systems can still cause collisions when faced with highly mobile traffic participants, such as two-wheeled vehicles suddenly appearing out of the way. Therefore, how intelligent vehicles can make path decisions and implement collision strategies that minimize occupant injury to reduce the incidence of two-wheeled vehicle accidents and the resulting injuries and fatalities has become a key research area in traffic safety.

[0003] Currently, most research in the field of intelligent vehicle path decision-making focuses on avoiding collisions and improving vehicle safety, traffic efficiency, and passenger comfort, but fails to address the issue of how to reduce injuries to people in dangerous situations where collisions are unavoidable.

[0004] Therefore, there is an urgent need for a method for intelligent vehicle hazard path decision-making for two-wheeled vehicle collision protection, which can make hazard path decisions for intelligent vehicles to solve the problem of how to reduce injury to people in hazard situations where collisions are unavoidable. Summary of the Invention

[0005] The present invention aims to provide a method for intelligent vehicle hazard path decision-making for collision protection of two-wheeled vehicles, which can make hazard path decisions for intelligent vehicles to solve the problem of how to reduce injury to people in hazard situations where collision is unavoidable.

[0006] This invention provides the following basic solution: a method for intelligent vehicle hazard path decision-making for two-wheeled vehicle collision protection, comprising the following:

[0007] S1. Acquire real-time images of the left and right visual systems, and construct a trajectory prediction module based on the real-time images of the left and right visual systems to predict the trajectory of the two-wheeled vehicle and generate the trajectory prediction result of the two-wheeled vehicle.

[0008] S2. Based on the trajectory prediction results of the two-wheeled vehicle, perform personnel injury prediction and generate personnel injury prediction results;

[0009] S3. Based on the personnel injury prediction results, make intelligent vehicle risk-based path decisions.

[0010] Furthermore, S101, based on publicly available autonomous driving datasets, traffic flow scene data containing two-wheeled vehicles is extracted; the features of the traffic flow scene data include: the position, speed, and acceleration of the two-wheeled vehicles and surrounding vehicles;

[0011] S102. Preprocess the traffic flow scenario data to obtain the preprocessed data;

[0012] S103. Construct an environment perception model based on deep learning to extract traffic flow scene data containing features from video images, and realize real-time detection and tracking of traffic participants.

[0013] S104. Using optical video sensors, fusing vehicle positioning and high-precision map information, and combining environmental perception models, historical trajectory data of the vehicle and surrounding traffic participants are obtained.

[0014] S105. Using an optical video sensor and an environmental perception model, a binocular stereo matching algorithm is used to extract key features of a specific target. The key features include the relative position, speed, and heading of the specific target.

[0015] S106. Use the K-nearest neighbor algorithm to preprocess and classify the historical trajectory data of traffic participants to obtain the historical trajectory data of different traffic participants;

[0016] S107. A coder-decoder trajectory prediction network framework is constructed using long short-term memory network, self-attention network, graph attention network and convolutional neural network as a trajectory prediction model. Based on the classified historical trajectory data, the trajectory prediction of two-wheeled vehicles is performed based on the multi-dimensional interactive characteristics of environmental state, participant behavior and rider driving intention.

[0017] Furthermore, the aforementioned method employs Long Short-Term Memory (LSTM) networks, Self-Attention Networks, Graph Attention Networks, and Convolutional Neural Networks to construct an encoder-decoder trajectory prediction network framework as a trajectory prediction model, including:

[0018] The encoder uses a long short-term memory network and a self-attention network to process the classified historical trajectory data and obtain the hidden state of the historical trajectory data.

[0019] Based on the hidden state of historical trajectory data, a graph attention network is used to capture the social interaction information between two-wheeled vehicles and surrounding vehicles.

[0020] A convolutional neural network is used to extract high-dimensional scene features from static map information;

[0021] Vector concatenation is performed on high-dimensional scene features and social interaction information;

[0022] A self-attention model decoder using a self-attention network decodes the concatenated vectors to obtain the future trajectory coordinates, which are then used as the trajectory prediction result for the two-wheeled vehicle.

[0023] Furthermore, the training process of the encoder-decoder trajectory prediction network framework is as follows:

[0024] The categorized historical trajectory data is divided into historical and future trajectories based on preset nodes and ranges. Historical trajectory data before a node is considered as historical trajectory data, and historical trajectory data after a node is considered as future trajectory data.

[0025] The historical trajectory serves as the input to the Long Short-Term Memory network and the first self-attention model encoder. The output of the first self-attention model encoder is input to the graph neural network and the second self-attention model encoder. At the same time, the output of the first self-attention model encoder and the future trajectory serve as the input to the second self-attention model encoder.

[0026] Static map information is used as input to the convolutional neural network, and the output of the convolutional neural network is used as input to the first multilayer perceptron.

[0027] The high-latitude scene features are input into the first multilayer perceptron, and the output results and social interaction information are vector concatenated. The vector concatenation results are then input into the multilayer perceptron and processed by Gaussian distribution.

[0028] The output of the graph neural network and the output of the first multilayer perceptron are concatenated. The result of the vector concatenation is then input into the second multilayer perceptron and processed with a Gaussian distribution. Simultaneously, the result of the vector concatenation is convolved with the output of the second self-attention model encoder, and the result of the convolution is input into the third multilayer perceptron and processed with a Gaussian distribution.

[0029] The results of the two Gaussian distributions are convolved and input into a fourth multilayer perceptron. The output of the fourth multilayer perceptron is then input into a self-attention model decoder for decoding to obtain the predicted trajectory.

[0030] The predicted trajectory and the future trajectory are compared and analyzed in detail. If the difference is within the preset difference range, the trajectory prediction model training is complete. If the difference is not within the preset difference range, the trajectory prediction model training continues.

[0031] Furthermore, the self-attention model encoder:

[0032]

[0033] Where, x t These are input features. This represents the hidden state of the encoder at time step t;

[0034] The self-attention model decoder:

[0035]

[0036] Where c is the context vector from the encoder. Let y be the hidden state of the decoder at time step t.t The predicted output is the coordinates of the future trajectories of traffic participants (vehicles).

[0037] Furthermore, S2 includes:

[0038] S201. Based on the two-wheeled vehicle historical database, analyze the damage characteristics of different collision types and establish the internal characteristics between different collision types and damage mechanisms.

[0039] S202. Reconstruct the collision accident to restore the dynamic response information of the human body at the moment of collision, thereby accurately capturing the kinematic mechanism of the rider's injury; in this embodiment, Madymo software is used for collision accident reconstruction.

[0040] S203. Establish a simulation matrix and build a database of injuries to two-wheeled vehicle riders based on a large number of collision simulation experiments.

[0041] S204. Filter the data in the two-wheeled vehicle rider injury database and perform data preprocessing operations; the two-wheeled vehicle rider injury database contains collision speed, collision angle, collision location and corresponding injury values; the data preprocessing operation involves using the injury values ​​contained in the injury database.

[0042] S205. Build a rider collision injury prediction model based on decision tree and neural network, and train and validate the injury prediction model based on the two-wheeled rider injury database established in step S2-4.

[0043] S206. Based on the trajectory prediction model, obtain the potential collision speed, collision location and collision angle between the two-wheeled vehicle and the autonomous vehicle.

[0044] S207. Using the injury prediction model trained in S205, and taking the collision speed, collision position, and collision angle obtained in S206 as input parameters for the injury prediction model, the potential injury value of the cyclist is obtained as the injury prediction result.

[0045] Furthermore, S202 includes:

[0046] S20201. Based on the Madymo modeling environment, a vehicle model was built and the force-deformation curve was used to characterize the vehicle's contact characteristics.

[0047] S20202. Build a two-wheeled vehicle model and fix different structural components of the two-wheeled vehicle by connecting different types of hinges;

[0048] S20203. Use the GEBOD module to scale the human body model in the Madymo environment;

[0049] S20204. Determine the initial design variables that affect accident reconstruction and their value ranges, where the initial design variables include: vehicle collision speed, collision angle, and collision location;

[0050] S20205. Set the reconstruction goal, constraints, and initial variables respectively. The platform automatically reads the results and uses them to calculate the objective function and penalty function to determine the feasibility of the current solution, and then decides the direction of algorithm optimization.

[0051] S20206. When the position error is controlled within 20% and the simulation results are consistent with the relevant information in the accident record, the accident reconstruction can be ended; otherwise, change the parameters such as the number of initial variables, the range of values, and the maximum number of optimizations, and enter the next automatic simulation and optimization cycle until the result meets the termination condition.

[0052] S20207. When the number of iterations reaches the specified upper limit, analyze the kinematics and injury response of the cyclist in the global optimal solution.

[0053] Furthermore, the data preprocessing operation described in S204 specifically includes:

[0054] X stand =(X-mean(X))std(X)

[0055] Where X is the original data, X stand The data are preprocessed, with mean(X) representing the average and std(X) representing the standard deviation.

[0056] Furthermore, the specific calculation process of the decision tree model described in S205 is as follows:

[0057]

[0058] in, This represents the predicted damage value, where K is the total number of trees, F represents the tree model, and x represents the predicted damage value. i It is the i-th input sample of the model;

[0059] The specific calculation process of the neural network model described in S205 is as follows:

[0060] z (l+1) =W (l) a (l) +b (l)

[0061] Among them, z (l+1) W is the output value of layer l of the model. (l) b represents the model weights. (l) a is the model bias value. (l) These are the input values ​​for layer l of the model.

[0062] Furthermore, S3 includes:

[0063] S301, Obtain several candidate vehicle trajectories;

[0064] S302. Treat vehicle dynamics constraints as hard constraints;

[0065] S303. Build a collision detection module and use it as a hard constraint to minimize the occurrence of collisions between the vehicle and two-wheeled vehicles.

[0066] S304. Traverse all candidate vehicle trajectories, first filtering out those that do not meet vehicle dynamics constraints and those that will collide.

[0067] S305. The filtered vehicle trajectories are sorted according to the personnel injury prediction results obtained in S2 above, and the vehicle trajectory corresponding to the smallest personnel injury prediction result is taken as the final autonomous vehicle planning trajectory.

[0068] S306. When all vehicle trajectories will collide with two-wheeled vehicles, the vehicle trajectory that causes the least damage to the two-wheeled vehicle rider shall be selected based on the personnel injury prediction results.

[0069] The beneficial effects of this solution are as follows: By building an injury prediction model for two-wheeled vehicle riders, this solution introduces the injury indicators of vulnerable road users into the autonomous vehicle path planning system, thereby realizing the proactive protection of vulnerable road users by autonomous vehicles and improving the safety and advancement of the autonomous driving system.

[0070] Specifically, this solution constructs an encoder-decoder trajectory prediction network framework based on existing data and various analytical models to predict the trajectory of two-wheeled vehicles, taking into account the multi-dimensional interactive characteristics of environmental conditions, participant behavior, and rider driving intentions. Based on the predicted trajectory results, and combined with two-wheeled vehicle epidemiological analysis, machine learning and neural network methods are used to predict personal injury, obtaining personal injury predictions corresponding to various two-wheeled vehicle trajectory prediction results. Finally, based on the personal injury predictions, intelligent vehicle risk path decision-making is performed, selecting the path with the least damage, thereby solving the problem of how to reduce personal injury in risky situations where collisions are unavoidable.

[0071] In summary, this solution enables intelligent vehicles to make dangerous path decisions, thereby addressing the issue of how to reduce harm to occupants in dangerous situations where collisions are unavoidable. Attached Figure Description

[0072] Figure 1 This is a flowchart illustrating an embodiment of an intelligent vehicle hazard path decision-making method for collision protection of two-wheeled vehicles according to the present invention.

[0073] Figure 2 This is a schematic diagram of the encoder-decoder trajectory prediction network framework of an embodiment of an intelligent vehicle hazard path decision-making method for two-wheeled vehicle collision protection according to the present invention;

[0074] Figure 3 This is a schematic diagram illustrating the training process of the encoding-decoding trajectory prediction network framework in an embodiment of the intelligent vehicle hazardous path decision-making method for two-wheeled vehicle collision protection according to the present invention. Detailed Implementation

[0075] The following detailed description illustrates the specific implementation method:

[0076] The basic implementation examples are as follows: Figure 1 As shown: A method for intelligent vehicle hazard path decision-making for two-wheeled vehicle collision protection, including the following:

[0077] S1. Acquire real-time images of the left and right visual systems, and construct a trajectory prediction model based on the real-time images of the left and right visual systems to predict the trajectory of the two-wheeled vehicle and generate the trajectory prediction result of the two-wheeled vehicle.

[0078] The predicted trajectory results for two-wheeled vehicles include: collision location, collision speed, and collision heading;

[0079] Specifically, S1 includes:

[0080] S101. Based on publicly available autonomous driving datasets, extract traffic flow scene data containing two-wheeled vehicles; the features of the traffic flow scene data include: the position, speed, acceleration, and other information of the two-wheeled vehicles and surrounding vehicles; the autonomous driving dataset is a collection of publicly available video images of autonomous driving.

[0081] S102. Preprocess the traffic flow scenario data to obtain preprocessed data; the preprocessing includes cleaning, noise reduction and standardization to ensure data quality.

[0082] S103. Construct an environment perception model based on deep learning to extract traffic flow scene data containing features from video images, and realize real-time detection and tracking of traffic participants.

[0083] Specifically, in this embodiment, the environmental perception model adopts a BP neural network. A training set is constructed using preprocessed traffic flow scene data and corresponding autonomous driving data. The training set is used to train the environmental perception model, enabling the environmental perception model to analyze and process the input video images and extract traffic flow scene data containing features.

[0084] S104. Using optical video sensors, fusing vehicle positioning and high-precision map information, and combining environmental perception models, historical trajectory data of the vehicle and surrounding traffic participants are obtained.

[0085] Specifically, an optical video sensor is used to acquire video images; the video images are real-time images of the left and right visual systems.

[0086] The environmental perception model extracts traffic flow scene data containing features from video images;

[0087] Based on traffic flow scenario data, the system integrates vehicle positioning and high-precision map information to analyze the trajectory data of the vehicle and surrounding traffic participants within a preset time period, which is then used as historical trajectory data. The preset time period refers to past time, hence the trajectory data is historical trajectory data. The length of the preset time period is set according to requirements. Specifically, based on the position, speed, and acceleration of the two-wheeled vehicle and surrounding vehicles, combined with vehicle positioning and high-precision map information, the system updates the trajectories of all traffic participants within the preset time period on the map, thereby obtaining historical trajectory data.

[0088] S105. Using an optical video sensor and an environmental perception model, a binocular stereo matching algorithm is used to extract key features of a specific target. The key features include the relative position, speed, and heading of the specific target.

[0089] Specifically, an optical video sensor is used to acquire video images; the video images are real-time images of the left and right visual systems.

[0090] The environmental perception model extracts traffic flow scene data containing key features for specific targets based on video images;

[0091] S106. The K-Nearest Neighbor (KNN) algorithm is used to preprocess and classify the historical trajectory data of traffic participants to obtain the historical trajectory data of different traffic participants in order to distinguish the traffic participants corresponding to the historical trajectory data.

[0092] S107. Using long short-term memory network, self-attention network, graph attention network and convolutional neural network, an encoder-decoder trajectory prediction network framework is constructed as a trajectory prediction model. Based on the classified historical trajectory data, the trajectory prediction of two-wheeled vehicles is performed based on the multi-dimensional interactive characteristics of environmental state, participant behavior and rider driving intention.

[0093] Specifically, it employs Long Short-Term Memory (LSTM) networks, Self-Attention Networks, Graph Attention Networks, and Convolutional Neural Networks to construct an encoder-decoder trajectory prediction network framework as the trajectory prediction model, such as... Figure 2 As shown: Long Short-Term Memory (LSTM) network and Self-Attention Network (Transformer) are used to process the classified historical trajectory data to obtain the hidden state of the historical trajectory data;

[0094] The specific structure of the Long Short-Term Memory network is as follows:

[0095] Forgotten Gate:

[0096] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0097] Among them, f t W represents the activation vector of the forget gate. f and b f These represent the weights and biases, respectively, where σ is the sigmoid activation function, and h... t-1 The hidden state of the previous time step, x t The current input is the categorized historical trajectory data;

[0098] Input Gate:

[0099] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0100]

[0101] Among them, i t It is the activation vector of the input gate. It is the candidate cell state;

[0102] Update cell status:

[0103]

[0104] Output gate:

[0105] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0106] h t =o t *tanh(C t );

[0107] The specific structure of the self-attention network is as follows:

[0108]

[0109] Where Q, K, and V are the query, key, and value matrices, respectively, and d k The dimension of the key;

[0110] Output the hidden state of historical trajectory data with higher dimensionality;

[0111] Specifically, the self-attention model encoder:

[0112]

[0113] Where, x t The input feature is historical trajectory data; This represents the hidden state of the encoder at time step t;

[0114] Based on the hidden state of historical trajectory data, a graph attention network (GAT) is used to capture the social interaction information between two-wheeled vehicles and surrounding vehicles; where social interaction information refers to the mutual influence relationship between different traffic participants (cars, two-wheeled vehicles, etc.), that is, the hidden state of historical trajectory data after the attention mechanism is processed.

[0115] The input of the graph attention network is the hidden state of the historical trajectory data, and the output is the hidden state of the historical trajectory data after the attention mechanism is processed.

[0116] The specific structure of the graph attention network is as follows:

[0117] Calculation of attention coefficient:

[0118]

[0119] Among them, e ij The initial values ​​for the attention coefficients between node i and node j; is a learnable weight vector used to calculate attention coefficients; W is a weight matrix used to linearly transform the features of nodes. and are the feature vectors of node i and node j, respectively; LeakyReLU is the activation function used to add non-linearity;

[0120] Normalized attention coefficient:

[0121]

[0122] Where, α ij The normalized attention coefficients indicate the importance of node j to node i. Let i be the set of neighboring nodes of node i;

[0123] Node feature update:

[0124]

[0125] in, σ is the updated feature vector of node i; σ is a non-linear activation function, typically using functions such as sigmoid or ReLU; in this embodiment, the ReLU function is used.

[0126] A convolutional neural network (CNN) is used to extract high-dimensional scene features from static map information.

[0127] Specifically, the input to a convolutional neural network is static map information, and the output is high-dimensional scene features after convolution processing.

[0128] The static map information can be a high-precision map or a simple map that only contains lane line information;

[0129] The specific structure of the convolutional neural network is as follows:

[0130]

[0131] Among them, W m b is the convolution kernel; f is the bias; f is the non-linear activation function; * denotes the convolution operation; x m The input is static map information, and C(x) is the output high-latitude scene features;

[0132] Vector concatenation is performed on high-dimensional scene features and social interaction information;

[0133] A self-attention model decoder is used to decode the concatenated vectors and obtain the future trajectory coordinates, i.e., the predicted trajectory, as the trajectory prediction result of the two-wheeled vehicle.

[0134] Self-attention model decoder:

[0135]

[0136] Where c is the context vector from the encoder. Let y be the hidden state of the decoder at time step t. t The predicted output is the coordinates of the future trajectories of traffic participants (vehicles).

[0137] The process of training an encoder-decoder trajectory prediction network framework, such as Figure 3 As shown:

[0138] The categorized historical trajectory data is divided into historical and future trajectories based on preset nodes and ranges. Historical trajectory data before a node is considered as historical trajectory data, and historical trajectory data after a node is considered as future trajectory data.

[0139] The historical trajectory serves as the input to the Long Short-Term Memory network and the first self-attention model encoder. The output of the first self-attention model encoder (the hidden state of the historical trajectory data) is input to the graph neural network and the second self-attention model encoder. At the same time, the output of the first self-attention model encoder and the future trajectory serve as the input to the second self-attention model encoder.

[0140] Static map information is used as input to the convolutional neural network, and the output of the convolutional neural network (high-latitude scene features) is input to the first multilayer perceptron.

[0141] The high-latitude scene features are input into the first multilayer perceptron, and the output results and social interaction information are vector concatenated. The vector concatenation results are then input into the multilayer perceptron and processed by Gaussian distribution.

[0142] The output of the graph neural network (social interaction information) and the output of the first multilayer perceptron are concatenated. The concatenated result is then input into the second multilayer perceptron and processed with a Gaussian distribution. Simultaneously, the concatenated result is convolved with the output of the second self-attention model encoder (the hidden state of the future trajectory). The convolution result is then input into the third multilayer perceptron and processed with a Gaussian distribution.

[0143] The results of the two Gaussian distributions are convolved and input into a fourth multilayer perceptron. The output of the fourth multilayer perceptron is then input into a self-attention model decoder for decoding to obtain the predicted trajectory.

[0144] The predicted trajectory and the future trajectory are compared and analyzed in detail. If the difference is within the preset difference range, the trajectory prediction model training is complete. If the difference is not within the preset difference range, the trajectory prediction model training continues.

[0145] S2. Based on the trajectory prediction results of the two-wheeled vehicle, perform personnel injury prediction and generate personnel injury prediction results;

[0146] The S2 method for predicting personal injury includes epidemiological analysis of two-wheeled vehicles, the establishment of an intelligent accident reconstruction platform, and the use of machine learning and neural network methods to predict personal injury. Its overall principle includes:

[0147] Specifically, S2 includes:

[0148] S201. Based on the two-wheeled vehicle-vehicle historical database, analyze the damage characteristics of different collision types and establish the in-line characteristics between different collision types and damage mechanisms. The two-wheeled vehicle-vehicle historical database can be a publicly available database or a database built according to requirements. The in-line characteristics refer to the correlation characteristics between different collision types between vehicles and two-wheeled vehicles and different injury types of two-wheeled vehicle riders.

[0149] S202. Reconstruct the collision accident to restore the dynamic response information of the human body at the moment of collision, so as to accurately capture the kinematic mechanism of the rider's injury. In this embodiment, the Madymo software is used to reconstruct the collision accident. The collision speed, collision position and collision angle at the moment of collision between the two-wheeled vehicle and the car are input into the Madymo software to obtain the dynamic response information of the human body.

[0150] Specifically, S202 includes:

[0151] S20201. Based on the Madymo modeling environment, build a vehicle model and use force-deformation curves to characterize the vehicle's contact characteristics; the vehicle model is the model of the vehicle that will collide with the two-wheeled vehicle.

[0152] S20202. Construct a two-wheeled vehicle model and fix different structural components of the two-wheeled vehicle by connecting them with different types of hinges; the two-wheeled vehicle model is the model of the two-wheeled vehicle that will be collided with.

[0153] S20203. The GEBOD module in Madymo is used to scale the human body model in the Madymo environment to meet the human characteristics of cyclists in real collision accidents.

[0154] S20204. Determine the initial design variables and their value ranges that affect accident reconstruction, wherein the initial design variables include: vehicle collision speed, collision angle, and collision position; in this embodiment, the initial design variables and their value ranges that affect accident reconstruction are determined in the intelligent accident reconstruction platform.

[0155] S20205. The reconstruction objective, constraints, and initial variables are set respectively. The intelligent accident reconstruction platform automatically reads the results and uses them to calculate the objective function and penalty function to determine the feasibility of the current solution, thereby determining the direction of algorithm optimization. This embodiment is performed within the intelligent accident reconstruction platform. The reconstruction objective is to restore the real collision scene; the constraints are the range of values ​​for the variables to be solved; the initial variables are collision speed, collision angle, and collision position; the objective function is the function that needs to be minimized or maximized during the optimization process. In accident reconstruction, the objective function is used to measure the difference between the reconstruction result and the actual accident situation. For example, the objective function is a measure of the error between the actual accident data and the reconstruction result, such as rider injury. The penalty function is used to penalize solutions that do not meet the constraints, guiding the optimization process towards solutions that meet the constraints. In accident reconstruction, the optimal solution is usually determined by minimizing the objective function; that is, the optimization algorithm will find the solution that minimizes the objective function value. Commonly used optimization algorithms in accident reconstruction include: genetic algorithm, particle swarm optimization algorithm, simulated annealing algorithm, and gradient descent algorithm.

[0156] S20206. When the position error is controlled within 20% and the simulation results are consistent with the relevant information in the accident record, the accident reconstruction can be ended; otherwise, change the parameters such as the number of initial variables, the range of values, and the maximum number of optimizations, and enter the next automatic simulation and optimization cycle until the result meets the termination condition.

[0157] S20207. When the number of iterations reaches the specified upper limit, analyze the kinematics and injury response of the cyclist in the global optimal solution; specifically, by comparing the cyclist's injury in the simulation with the cyclist's injury in the actual collision accident, analyze the kinematics and injury response of the cyclist in the global optimal solution.

[0158] S203. Establish a simulation matrix and build a two-wheeled vehicle rider injury database based on a large number of collision simulation experiments. The simulation matrix includes different collision speeds, collision positions and collision angles to obtain a large amount of collision scene data and construct a two-wheeled vehicle rider injury database.

[0159] S204. Filter the data in the two-wheeled vehicle rider injury database and perform data preprocessing operations; the two-wheeled vehicle rider injury database contains collision speed, collision angle, collision location and corresponding injury values; the data preprocessing operation involves using the injury values ​​contained in the injury database.

[0160] The data preprocessing operation described in S204 is specifically as follows:

[0161] X stand =(X-mean(X))std(X)

[0162] Where X is the original data, X stand For the preprocessed data, mean(X) is the mean and std(X) is the standard deviation;

[0163] S205. Build a rider collision injury prediction model based on decision tree and neural network, and train and validate the injury prediction model based on the two-wheeled rider injury database.

[0164] The specific calculation process of the decision tree model described in S205 is as follows:

[0165]

[0166] in, This represents the predicted damage value, i.e., the damage value. K is the total number of trees, F represents the tree model, and x i It is the i-th input sample of the model, namely the collision speed, collision angle, and collision location in the two-wheeled rider injury database;

[0167] The specific calculation process of the neural network model described in S205 is as follows:

[0168] z (l+1) =W (l) a (l) +b (l)

[0169] Among them, z (l+1) W is the output value of layer l of the model. (l) b represents the model weights. (l) a is the model bias value. (l) The input values ​​are for layer l of the model; the input values ​​are the collision speed, collision angle, and collision location from the two-wheeled rider injury database; the output value is the injury value.

[0170] S206. Based on the trajectory prediction model, obtain the potential collision speed, collision location and collision angle between the two-wheeled vehicle and the autonomous vehicle.

[0171] S207. Using the injury prediction model trained in S205, and taking the collision speed, collision position, and collision angle obtained in S206 as input parameters for the injury prediction model, the potential injury risk of the cyclist, i.e., the injury value, is obtained as the injury prediction result.

[0172] S3. Based on the personnel injury prediction results, make intelligent vehicle risk status path decisions;

[0173] Specifically, S3 includes:

[0174] S301. Obtain several candidate vehicle trajectories; In this embodiment, a series of candidate vehicle trajectories are obtained by sampling within the drivable area of ​​the vehicle.

[0175] S302. Vehicle dynamics constraints are treated as hard constraints; vehicle dynamics constraints include: speed constraints, acceleration constraints, and steering constraints.

[0176] S303. Build a collision detection module, which, together with vehicle dynamics constraints, serves as a hard constraint to minimize the occurrence of collisions between the vehicle and two-wheeled vehicles.

[0177] S304. Traverse all candidate vehicle trajectories, first filtering out those that do not meet vehicle dynamics constraints and those that will collide.

[0178] S305. The filtered vehicle trajectories are sorted according to the personnel injury prediction results obtained in S2 above, and the vehicle trajectory corresponding to the smallest personnel injury prediction result is taken as the final autonomous vehicle planning trajectory.

[0179] S306. When all vehicle trajectories will collide with two-wheeled vehicles, the vehicle trajectory that causes the least damage to the two-wheeled vehicle rider shall be selected based on the personnel injury prediction results.

[0180] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for intelligent vehicle hazard path decision-making for two-wheeled vehicle collision protection, characterized in that, Includes the following: S1. Acquire real-time left and right visual images, and based on the real-time left and right visual images, construct a trajectory prediction model to predict the trajectory of the two-wheeled vehicle and generate the trajectory prediction result; S1 includes: S101. Based on publicly available autonomous driving datasets, extract traffic flow scene data containing two-wheeled vehicles; the features of the traffic flow scene data include: the position, speed, and acceleration of the two-wheeled vehicles and surrounding vehicles. S102. Preprocess the traffic flow scenario data to obtain the preprocessed data; S103. Construct an environment perception model based on deep learning to extract traffic flow scene data containing features from video images, and realize real-time detection and tracking of traffic participants. S104. Using optical video sensors, fusing vehicle positioning and high-precision map information, and combining environmental perception models, historical trajectory data of the vehicle and surrounding traffic participants are obtained. S105. Using an optical video sensor and an environmental perception model, a binocular stereo matching algorithm is used to extract key features of a specific target. The key features include the relative position, speed, and heading of the specific target. S106. Preprocess and classify the historical trajectory data of traffic participants to obtain historical trajectory data of different traffic participants; S107. Using long short-term memory network, self-attention network, graph attention network and convolutional neural network, an encoder-decoder trajectory prediction network framework is constructed as a trajectory prediction model. Based on the classified historical trajectory data, the trajectory prediction of two-wheeled vehicles is performed based on the multi-dimensional interactive characteristics of environmental state, participant behavior and rider driving intention. S2. Based on the trajectory prediction results of the two-wheeled vehicle, perform personnel injury prediction and generate personnel injury prediction results; S3. Based on the personnel injury prediction results, make intelligent vehicle risk-based path decisions.

2. The intelligent vehicle hazard path decision-making method for two-wheeled vehicle collision protection according to claim 1, characterized in that, The aforementioned approach employs Long Short-Term Memory (LSTM) networks, Self-Attention Networks, Graph Attention Networks, and Convolutional Neural Networks to construct an encoder-decoder trajectory prediction network framework as the trajectory prediction model, including: The encoder uses a long short-term memory network and a self-attention network to process the classified historical trajectory data and obtain the hidden state of the historical trajectory data. Based on the hidden state of historical trajectory data, a graph attention network is used to capture the social interaction information between two-wheeled vehicles and surrounding vehicles. A convolutional neural network is used to extract high-dimensional scene features from static map information; Vector concatenation is performed on high-dimensional scene features and social interaction information; A self-attention model decoder using a self-attention network decodes the concatenated vectors to obtain the future trajectory coordinates, which are then used as the trajectory prediction result for the two-wheeled vehicle.

3. The intelligent vehicle hazard path decision-making method for two-wheeled vehicle collision protection according to claim 2, characterized in that, The training process of the encoder-decode trajectory prediction network framework is as follows: The categorized historical trajectory data is divided into historical and future trajectories based on preset nodes and ranges. Historical trajectory data before a node is considered as historical trajectory data, and historical trajectory data after a node is considered as future trajectory data. The historical trajectory serves as the input to the Long Short-Term Memory network and the first self-attention model encoder. The output of the first self-attention model encoder is input to the graph neural network and the second self-attention model encoder. At the same time, the output of the first self-attention model encoder and the future trajectory serve as the input to the second self-attention model encoder. Static map information is used as input to the convolutional neural network, and the output of the convolutional neural network is used as input to the first multilayer perceptron. The high-latitude scene features are input into the first multilayer perceptron, and the output results and social interaction information are vector concatenated. The vector concatenation results are then input into the multilayer perceptron and processed by Gaussian distribution. The output of the graph neural network and the output of the first multilayer perceptron are concatenated. The result of the vector concatenation is then input into the second multilayer perceptron and processed with a Gaussian distribution. Simultaneously, the result of the vector concatenation is convolved with the output of the second self-attention model encoder, and the result of the convolution is input into the third multilayer perceptron and processed with a Gaussian distribution. The results of the two Gaussian distributions are convolved and input into a fourth multilayer perceptron. The output of the fourth multilayer perceptron is then input into a self-attention model decoder for decoding to obtain the predicted trajectory. The predicted trajectory and the future trajectory are compared and analyzed in detail. If the difference is within the preset difference range, the trajectory prediction model training is complete. If the difference is not within the preset difference range, the trajectory prediction model training continues.

4. The intelligent vehicle hazard path decision-making method for two-wheeled vehicle collision protection according to claim 3, characterized in that, The self-attention model encoder: ； in, These are input features. For the encoder at time step Hidden state; The self-attention model decoder: ； in, The context vector comes from the encoder. For the decoder at time step The hidden state, The predicted output is the coordinates of the future trajectory of the traffic participants.

5. The intelligent vehicle hazard path decision-making method for two-wheeled vehicle collision protection according to claim 4, characterized in that, S2 includes: S201. Based on the two-wheeled vehicle historical database, analyze the damage characteristics of different collision types and establish the internal characteristics between different collision types and damage mechanisms. S202. Reconstruct the collision accident and restore the dynamic response information of the human body at the moment of collision, so as to accurately capture the kinematic mechanism of the rider's injury. S203. Establish a simulation matrix and build a database of injuries to two-wheeled vehicle riders based on a large number of collision simulation experiments. S204. Filter the data in the two-wheeled vehicle rider injury database and perform data preprocessing operations; the two-wheeled vehicle rider injury database contains collision speed, collision angle, collision location and corresponding injury values; the data preprocessing operation involves using the injury values ​​contained in the injury database. S205. Build a rider collision injury prediction model based on decision tree and neural network, and train and validate the injury prediction model based on the two-wheeled rider injury database established in step S2-4. S206. Based on the trajectory prediction model, obtain the potential collision speed, collision location and collision angle between the two-wheeled vehicle and the autonomous vehicle. S207. Using the injury prediction model trained in S205, and taking the collision speed, collision position, and collision angle obtained in S206 as input parameters for the injury prediction model, the potential injury value of the cyclist is obtained as the injury prediction result.

6. The intelligent vehicle hazard path decision-making method for two-wheeled vehicle collision protection according to claim 5, characterized in that, S202 includes: S20201. Based on the Madymo modeling environment, a vehicle model was built and the force-deformation curve was used to characterize the vehicle's contact characteristics. S20202. Build a two-wheeled vehicle model and fix different structural components of the two-wheeled vehicle by connecting different types of hinges; S20203. Use the GEBOD module to scale the human body model in the Madymo environment; S20204. Determine the initial design variables that affect accident reconstruction and their value ranges, where the initial design variables include: vehicle collision speed, collision angle, and collision location; S20205. Set the reconstruction goal, constraints, and initial variables respectively. The platform automatically reads the results and uses them to calculate the objective function and penalty function to determine the feasibility of the current solution, and then decides the direction of algorithm optimization. S20206. When the position error is controlled within 20% and the simulation results are consistent with the relevant information in the accident record, the accident reconstruction can be terminated. Otherwise, change the parameters such as the number of initial variables, the range of values, and the maximum number of optimizations, and enter the next automatic simulation and optimization cycle until the result meets the termination condition. S20207. When the number of iterations reaches the specified upper limit, analyze the kinematics and injury response of the cyclist in the global optimal solution.

7. The intelligent vehicle hazard path decision-making method for two-wheeled vehicle collision protection according to claim 6, characterized in that, The data preprocessing operation described in S204 is specifically as follows: Where X represents the original data. For the preprocessed data, This is the average value. The standard deviation is denoted as .

8. The intelligent vehicle hazard path decision-making method for two-wheeled vehicle collision protection according to claim 7, characterized in that, The specific calculation process of the decision tree model described in S205 is as follows: in, Indicates the predicted damage value. K It is the total number of trees. F Represents a tree model. It is the first of the models One input sample; The specific calculation process of the neural network model described in S205 is as follows: in, For the model l The output value of the layer, W (l) b represents the model weights. (l) a is the model bias value. (l) For the model l The input values ​​of the layer.

9. The intelligent vehicle hazard path decision-making method for two-wheeled vehicle collision protection according to claim 8, characterized in that, The S3 includes: S301, Obtain several candidate vehicle trajectories; S302. Treat vehicle dynamics constraints as hard constraints; S303. Build a collision detection module and use it as a hard constraint to minimize the occurrence of collisions between the vehicle and two-wheeled vehicles. S304. Traverse all candidate vehicle trajectories, first filtering out those that do not meet vehicle dynamics constraints and those that will collide. S305. The filtered vehicle trajectories are sorted according to the personnel injury prediction results obtained in S2 above, and the vehicle trajectory corresponding to the smallest personnel injury prediction result is taken as the final autonomous vehicle planning trajectory. S306. When all vehicle trajectories will collide with two-wheeled vehicles, the vehicle trajectory that causes the least damage to the two-wheeled vehicle rider shall be selected based on the personnel injury prediction results.

Images