Collision risk prediction method for self-driving automobile aiming at long-tail phenomenon

Abstract

The invention belongs to the field of road vehicle control, and relates to an automatic driving vehicle collision risk prediction method for a long-tail phenomenon, which comprises the following steps of: screening high-risk track samples, and realizing high-fidelity modeling and synthesis of rare high-risk driving behaviors by adopting a track generation method fusing a diffusion probability model and a Transformer structure. Therefore, the learning ability of the model to the long-tail sample is enhanced; besides, in consideration of the fact that in a long-tail scene, the kinematics characteristics and the traffic flow density characteristics of the vehicles generally have significant fluctuations, the kinematics characteristics and the traffic flow density characteristics of the surrounding traffic participants are constructed and extracted, and the future trajectories of the surrounding vehicles are predicted through multi-modal characteristic fusion; on the basis, a probability risk assessment mechanism for different risk scenes is constructed, the potential collision risk or behavior anomaly probability is dynamically assessed, and finally refined prediction of potential high-risk events in the complex traffic environment is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of road vehicle control, and relates to predicting or avoiding possible or upcoming collisions. Specifically, it relates to a method for predicting the collision risk of autonomous vehicles for the long-tail phenomenon. Background Art

[0002] With the rapid development of autonomous driving technology, vehicle driving safety has become one of the key prerequisites for the implementation of technology applications. Autonomous vehicles perceive and predict the surrounding traffic environment in real time to identify and avoid possible collision risks in advance, thereby ensuring driving safety. However, the existing methods for predicting the risks of autonomous vehicles still have significant deficiencies, especially when dealing with the "long-tail phenomenon" in real traffic scenarios. The long-tail phenomenon refers to a small number of extremely dangerous traffic scenarios (such as sudden lane changes, hard braking, sudden collision scenarios, etc.). Due to the extremely scarce data of extremely dangerous scenarios, while the data of most scenarios (such as normal and stable driving) is relatively rich, this imbalance in data distribution causes existing prediction models to easily ignore or underestimate the occurrence probability and severity of these extremely dangerous scenarios, resulting in problems such as excessive prediction errors for high-risk trajectories, response lags, or risk identification failures. Therefore, there is an urgent need to design a new collision risk prediction method to accurately predict the long-tail phenomenon, so as to improve the risk identification ability and prediction accuracy of the autonomous driving system for potential risks in complex scenarios. Summary of the Invention

[0003] In view of the above technical problems and deficiencies, the present invention proposes a method for predicting the collision risk of autonomous vehicles for the long-tail phenomenon. This method first processes the operation data collected by autonomous vehicles, and focuses on screening high-risk trajectory samples among them. On this basis, a trajectory generation method that combines a diffusion probability model and a Transformer structure is adopted to achieve high-fidelity modeling and synthesis of rare high-risk driving behaviors, so as to make up for the lack of data of rare dangerous events in long-tail scenarios, thereby enhancing the model's learning ability for long-tail samples. In addition, considering that in long-tail scenarios, the kinematic characteristics of vehicles (such as speed, acceleration, etc.) and traffic flow density characteristics (such as local traffic flow density, lane change frequency, etc.) usually show significant fluctuations, the present invention further constructs and extracts the kinematic characteristics and traffic flow density characteristics of surrounding traffic participants, and predicts the future trajectories of surrounding vehicles through multi-modal feature fusion. On this basis, a probability risk assessment mechanism for different risk scenarios is constructed to dynamically evaluate the potential collision risk or abnormal behavior probability, and finally achieve refined prediction of potential high-risk events in complex traffic environments.

[0004] To achieve the above object, the present invention adopts the following technical solutions: A method for predicting the collision risk of autonomous vehicles against the long-tail phenomenon, which uses a multi-modal enhanced trajectory dataset to train a collision risk prediction model, and then uses the trained collision risk prediction model to perform risk prediction. The specific risk prediction steps are as follows: Step 1. Construct the dynamic feature vector of surrounding vehicles; Step 2. Output the dynamic fluctuation type according to the dynamic feature vector of surrounding vehicles; Step 3. Construct the traffic flow feature vector; Step 4. Output the traffic flow state according to the traffic flow feature vector; Step 5. Trajectory prediction with multi-modal feature fusion: Encode the historical trajectory information of surrounding vehicles to obtain historical trajectory features , encode the high-precision map information to obtain map features ; perform feature mapping on the dynamic feature vector and traffic flow feature vector of surrounding vehicles to obtain dynamic features and traffic flow features ; splice the historical trajectory features and dynamic features to obtain features , splice the map features and traffic flow features to obtain features ; perform interactive modeling on the features and features to obtain features ; process the features to obtain features ; decode the features to decode the future trajectories, speeds, and accelerations of surrounding vehicles; Step 6. Collision risk prediction: Select the vehicle with the smallest distance from the autonomous vehicle as the target vehicle according to historical data, extract the maximum speed and maximum acceleration of the target vehicle, and calculate the standard deviation of acceleration; construct a dynamic collision risk index based on the maximum speed, maximum acceleration, and standard deviation of acceleration of the target vehicle; finally, determine the dynamic fluctuation coefficient according to the dynamic fluctuation type, determine the traffic flow state label according to the traffic flow state, and use the dynamic fluctuation coefficient and traffic flow state label to correct the dynamic collision risk index.

[0005] As a preference of the present invention, the construction steps of the multi-modal enhanced trajectory dataset are as follows: Step A. Data collection and feature construction: Step A1. Screen the trajectory data and construct the tail trajectory data sample; Step A2. Standardize the filtered tail trajectory data; Step A3. Extract trajectory features; Extract multi-dimensional features of the target vehicle from the standardized tail trajectory data ; where P represents the position sequence of the target vehicle, V represents the speed sequence of the target vehicle, A represents the acceleration sequence of the target vehicle, and K represents the curvature sequence of the target vehicle; Step B. Data augmentation based on the diffusion model: Step B1. Based on the constructed multi-dimensional features of the target vehicle , add Gaussian noise at each time step , to obtain the feature ; Step B2. Use the Transformer neural network to predict the Gaussian noise added at each time step , and then denoise the feature of each time step feature to obtain the feature ; Step B3. Input the denoised feature into a multi-layer perceptron to generate trajectory data; Step B4. By sampling different Gaussian noises multiple times , repeat the process of steps B2 to B3 to generate multiple different trajectory samples, thereby constructing a multi-modal enhanced trajectory dataset.

[0006] As a preference of the present invention, in step 1, calculate the speed mean , acceleration mean , speed standard deviation , speed peak , speed skewness , speed coefficient of variation based on the historical data of the surrounding vehicle i, and construct the dynamic feature vector of the surrounding vehicle i; Step 2. Compare the dynamic feature vector of the surrounding vehicle i with the cluster centers representing high dynamic fluctuations, medium dynamic fluctuations, and low dynamic fluctuations, and output the dynamic fluctuation type of the surrounding vehicle i.

[0007] As a preference of the present invention, in step 3, calculate the traffic density , traffic flow , and average speed of surrounding vehicles according to the sensing range radius R of the sensors on the autonomous vehicle and the number of vehicles N within the sensing range, and construct the traffic flow feature vector ={}; Step 4 compares the traffic flow feature vector around the autonomous vehicle with the cluster centers representing high density, medium density, and low density and outputs the traffic flow state.

[0008] As a preference of the present invention, in step 5, the LSTM encoder is used to encode the historical trajectory information to obtain the historical trajectory features , and the point Net-based encoder is used to encode the high-precision map information to obtain the map features ; the multi-layer perceptron is used to perform feature mapping on the dynamic feature vector and traffic flow feature vector of surrounding vehicles to obtain the dynamic features of surrounding vehicles and traffic flow features .

[0009] As a preference of the present invention, in step 5, the features and the features are input into the multi-head cross-attention module for interactive modeling to obtain the features ; the feed-forward neural network is used to process the features to obtain the features ; the multi-layer perceptron is used to decode the features .

[0010] As a preference of the present invention, the specific steps of the collision risk prediction in step 6 are as follows: Step 6.1. Calculate the distances from N vehicles around at the th moment in the historical time step to the autonomous vehicle, and select the vehicle with the smallest distance from the autonomous vehicle as the target vehicle; Step 6.2. According to the future speed and acceleration of the target vehicle, extract the maximum speed , maximum acceleration of the target vehicle; Step 6.3. Calculate the acceleration standard deviation according to the future acceleration of the target vehicle; Step 6.4. Use the maximum speed , maximum acceleration and acceleration standard deviation of the target vehicle to construct the dynamic collision risk index , and the expression is: ； In the formula, , are the weight coefficients of speed, acceleration, and fluctuation, represents the maximum speed limit of the road, For a safe acceleration, is the expected safe fluctuation value; Step 6.5. Set the dynamic fluctuation coefficient Correct the dynamic collision risk index , and obtain the dynamic collision risk index considering dynamic fluctuations , and the expression is: ; Step 6.6. Set the traffic flow state label For the dynamic collision risk index considering dynamic fluctuations Make corrections to obtain the final dynamic collision risk index , and the expression is: ; Step 6.7. Input the final dynamic risk index into the risk judgment module for judgment, and output the risk level .

[0011] As a further preference of the present invention, the speed peak value , speed skewness , and speed coefficient of variation in step 1 are expressed as: ; ; ; wherein, represents the speed at the f-th time step, is the time step of the historical trajectory.

[0012] As a further preference of the present invention, in step 6.4 , are respectively set to 1.0, 1.2, 1.0; is set to , is set to ; In step 6.5, high dynamic fluctuations , medium dynamic fluctuations low dynamic fluctuations ; In step 6.6, at high density , at medium density , at low density .

[0013] Advantages and beneficial effects of the present invention: (1) By screening high-risk trajectory data and combining a trajectory generation method based on a diffusion probability model and a Transformer structure, the present invention can synthesize rare high-risk scenarios with high fidelity, effectively expand the long-tail data samples, and improve the model's coverage ability and learning effect for extremely dangerous scenarios.

[0014] (2) The present invention constructs a high-dimensional input space that integrates historical trajectories, kinematic information, and traffic flow density features, enabling the model to more comprehensively understand the dynamic evolution process in complex traffic environments, and thus having strong prediction robustness and generalization ability in various potential risk scenarios.

[0015] (3) The present invention not only predicts the future trajectory of the target vehicle, but also introduces the behavior modeling and density perception of surrounding traffic participants. Combining the trajectory generation and risk discrimination modules, it constructs a probabilistic risk assessment mechanism for high-risk scenarios, which can more precisely identify and quantify potential collision risks.

[0016] (4) The method provided by the present invention is applicable to various scenarios, including urban congested sections, high-speed driving scenarios, and even when the sensor perception decreases in bad weather. It can stably perform risk prediction on traditional vehicles or improved autonomous driving platforms, providing good support for the implementation of autonomous driving in diverse environments.

[0017] (5) Compared with traditional rule-based or static model methods, the present invention can achieve earlier and more accurate risk warnings through intensive training of deep learning models on long-tail data, improve the reaction efficiency and decision-making quality of the system in sudden and dangerous scenarios, and enhance the overall driving safety.

[0018] (6) The method provided by the present invention is compatible with existing hardware conditions. By analyzing the environmental perception data collected by vehicle-mounted sensors and limited historical driving data, it does not require additional installation of special sensing devices or expensive hardware modules, has high economic efficiency and feasibility, and is convenient for integration and deployment with existing autonomous driving systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Through the following description with reference to the accompanying drawings, and with a more comprehensive understanding of the present invention, other objects and results of the present invention will become more apparent and easier to understand. In the drawings: Figure 1 is a flowchart of a method for predicting the collision risk of an autonomous vehicle for the long-tail phenomenon provided by the present invention; Figure 2 is a flowchart for constructing a multi-modal enhanced trajectory data set of the present invention; Figure 3 is a flowchart for constructing the dynamic characteristics of surrounding vehicles of the present invention; Figure 4Flow chart for constructing traffic flow characteristics of the present invention; Figure 5 Flow chart for trajectory prediction of multi-modal feature fusion of the present invention; Figure 6 Flow chart for collision risk prediction of the present invention. Detailed implementation manners

[0020] To enable those skilled in the art to better understand the technical solutions and advantages of the present invention, the present application will be described in detail below with reference to the accompanying drawings, but it is not intended to limit the protection scope of the present invention.

[0021] As Figure 1 shown, this embodiment provides a method for predicting the collision risk of an autonomous vehicle for the long-tail phenomenon. This method uses a multi-modal enhanced trajectory dataset to train a collision risk prediction model, and then uses the trained collision risk prediction model to perform risk prediction. The specific risk prediction steps are as follows: Step 1. Calculate the mean speed , mean acceleration , standard deviation of speed , peak speed , skewness of speed , coefficient of variation of speed based on the historical data of the surrounding vehicle i, and construct the dynamic feature vector of the surrounding vehicle i; Step 2. Compare the dynamic feature vector of the surrounding vehicle i with the cluster centers representing high dynamic fluctuations, medium dynamic fluctuations, and low dynamic fluctuations, and output the dynamic fluctuation type of the surrounding vehicle i; Step 3. Calculate the traffic density , traffic flow , and average speed of surrounding vehicles according to the sensing range radius R of the sensor on the autonomous vehicle and the number of vehicles N within the sensing range, and construct the traffic flow feature vector = { }; Step 4. Compare the traffic flow feature vector around the autonomous vehicle with the cluster centers representing high density, medium density, and low density, and output the traffic flow state; Step 5. Trajectory prediction of multi-modal feature fusion: Step 5.1. Vectorize and encode the historical trajectory information of surrounding vehicles and the high-precision map; Step 5.2. Encode the historical trajectory information of surrounding vehicles to obtain historical trajectory features , encode the high-precision map information to obtain map features ; Step 5.3. For the dynamic feature vectors of the surrounding N vehicles and traffic flow feature vectors ={ }, perform feature mapping to obtain the dynamic features and traffic flow features of the surrounding vehicles; Step 5.4. Concatenate the historical trajectory features and the dynamic features of the vehicle to obtain the feature , concatenate the map features and the traffic flow features to obtain the feature ; Step 5.5. Input the feature and the feature into the multi-head cross-attention module for interactive modeling to obtain the feature ; Step 5.6. Process the feature using a feed-forward neural network to obtain the feature ; Step 5.7. Decode the feature to decode the future trajectories of the surrounding N vehicles, the future speeds }, and the future accelerations ; Step 6. Collision risk prediction: Select the vehicle with the minimum distance from the autonomous vehicle as the target vehicle according to the historical data, and extract the maximum speed , maximum acceleration of the target vehicle according to the future speed and acceleration of the target vehicle, calculate the acceleration standard deviation , and construct a dynamic collision risk index based on the maximum speed , maximum acceleration and acceleration standard deviation of the target vehicle; finally, determine the dynamic fluctuation coefficient according to the type of dynamic fluctuation, determine the traffic flow state label according to the traffic flow state, and use the dynamic fluctuation coefficient , traffic flow state label to correct the dynamic collision risk index .

[0022] As Figure 2 shown, in this embodiment, the construction steps of the multi-modal enhanced trajectory dataset are as follows: Step A. Data collection and feature construction: Step A1. Screening of high-risk trajectory data; Collect the driving data of autonomous vehicles in the actual road traffic environment, and obtain the historical information of surrounding vehicles through on-vehicle sensors (such as lidar, millimeter-wave radar, cameras, GPS sensors, etc.); then preprocess the original data, screen out the trajectory segments containing obvious abnormal driving behaviors or high-collision-risk events, and use them as the tail trajectory data samples.

[0023] In this embodiment, the historical information of surrounding vehicles includes: information such as the trajectory information of surrounding vehicles, the types of surrounding vehicles, and the types of roads; among them, the trajectory information of surrounding vehicles includes the coordinates of the historical trajectory of the vehicle, the speed of the vehicle, the acceleration, the steering angle, etc.; the types of surrounding vehicles include bicycles, electric vehicles, cars, trucks, etc. Obvious abnormal driving behaviors or high-collision-risk events include but are not limited to: vehicle emergency braking events, vehicle sharp turning or sudden turning events, vehicle rapid lane change accompanied by obvious deceleration or acceleration events, vehicle spacing rapidly decreasing, and events where the collision risk index (such as time to collision TTC) is less than the set threshold.

[0024] Step A2. Standardize the screened tail trajectory data, and the standardization process includes time synchronization and coordinate system synchronization; specifically, when performing time synchronization, use the pulse signal provided by GPS to synchronize multiple on-vehicle sensors, and the coordinate system synchronization can refer to the existing method for coordinate system synchronization.

[0025] Step A3. Trajectory feature extraction; Extract the multi-dimensional features of the target vehicle from the standardized tail trajectory data ; where P represents the position sequence of the target vehicle, V represents the speed sequence of the target vehicle, A represents the acceleration sequence of the target vehicle, K represents the curvature sequence of the target vehicle, represents the total historical time step, represents the multi-dimensional feature at the t-th time step; The position sequence of the target vehicle ; where, represents the position at the t-th time step, ; The speed sequence of the target vehicle ; where, represents the speed at the t-th time step; The acceleration sequence of the target vehicle ; where, represents the acceleration at the t-th time step, and can be calculated by the following formula:

[0026] Among them, , are the speeds at two adjacent times (time step t + 1 and time step t), respectively; is the sampling time interval, set to 0.1 s; The sequence of the direction angles of the target vehicle ; where, represents the direction angle at the t-th time step; The sequence of the curvatures of the target vehicle is ; where, represents the curvature at the t-th time step and can be calculated using the following formula:

[0027] where, are the direction angles at two adjacent times (time step t + 1 and time step t), respectively, are the position coordinates at two adjacent times (time step t + 1 and time step t).

[0028] Step B. Data augmentation based on the Diffusion Model (DDPM): Step B1. Based on the constructed multi-dimensional features of the target vehicle , Gaussian noise is added at each time step to make the multi-dimensional features gradually degenerate and follow a random Gaussian distribution, obtaining the feature , , and the expression of the feature at the t-th time step after adding noise is:

[0029] In the formula, is the Gaussian noise; is the noise scheduling coefficient at time step t, that is, when time step t increases, , representing the cumulative signal retention rate from 1 to the t-th time step, represents the signal retention rate at the s-th time step, becomes smaller and smaller, resulting in becoming more and more randomized and losing the original features ； is the multi-dimensional feature at the t-th time step.

[0030] Step B2. Use the existing Transformer neural network to predict the Gaussian noise added at each time step , and then denoise each time step feature in the feature to obtain the feature , , the features at the t-th time step after denoising The expression is:

[0031]

[0032] In the formula, represents the "signal retention rate" of each time step .

[0033] Step B3. Input the above denoised features into the MLP (Multi-Layer Perceptron), and the multi-layer perceptron generates trajectory data with physical rationality and behavioral feature consistency. The expression of the trajectory is:

[0034] In the formula, the generated trajectory , represents the trajectory at the t-th time step. The MLP mainly consists of two linear transformation layers (fully connected layers) and a ReLU activation function.

[0035] Step B4. By sampling different Gaussian noises multiple times and repeating the process of Steps B2 to B3, multiple different trajectory samples can be generated, thereby constructing a multi-modal enhanced trajectory dataset.

[0036] As Figure 3 shown, in this embodiment, Step 1 specifically includes the following steps: Step 1.1. Calculate the velocity mean according to the velocity sequence of the surrounding vehicle i within the historical trajectory time steps ; Step 1.2. Calculate the acceleration mean according to the velocity sequence of the surrounding vehicle i within the historical trajectory time steps , and the expression is:

[0037] In the formula, represents the time interval of the sampled historical trajectory, , respectively represent the velocities at the f-th time step and the f + 1-th time step; Step 1.3. Calculate the velocity standard deviation according to the velocity sequence of the surrounding vehicle i and the velocity mean , and the expression is:

[0038] Among them, represents the speed at the f-th time step; Step 1.4. Calculate the speed peak , the average speed and the standard deviation of speed based on the speed sequence of the surrounding vehicle i , which is expressed as:

[0039] Step 1.5. Calculate the speed skewness , the average speed and the standard deviation of speed based on the speed sequence of the surrounding vehicle i , which is expressed as:

[0040] Step 1.6. Calculate the coefficient of variation of speed and the standard deviation of speed based on the average speed of vehicle i , which is expressed as:

[0041] Step 1.7. Concatenate the above calculated metrics to obtain the dynamic feature vector of the surrounding vehicle i .

[0042] In this embodiment, the dynamic feature vectors of all surrounding vehicles can be obtained in the above manner. The dynamic feature vectors of all surrounding vehicles are expressed as , where N represents the total number of vehicles; in actual operation, the dynamic feature vectors of vehicles in all the collected data are used as a data set, and K-means clustering is used to cluster out three cluster centers , and the three cluster centers respectively correspond to: high dynamic fluctuation, medium dynamic fluctuation, and low dynamic fluctuation; subsequently, the dynamic feature vector of the surrounding vehicle i is compared with the three cluster centers , and the dynamic fluctuation type of the surrounding vehicle i can be directly output.

[0043] As Figure 4 shown, in this embodiment, step 3 specifically includes the following steps: Step 3.1. Calculate the traffic density based on the sensing range radius R of the sensors of the autonomous vehicle and the number of vehicles N in the sensing area at time t

[0044] In the formula, represents the pi, and R represents the radius of the sensing range.

[0045] Step 3.2. Calculate the traffic flow according to the number of vehicles within the sensing range at moment and the number of vehicles N within the sensing area at time t , and the expression is: :

[0046] In the formula, represents the absolute value.

[0047] Step 3.3. Calculate the average speed according to the speeds of N vehicle numbers within the sensing area at time t }, and the expression is: :

[0048] Step 3.4. Concatenate to obtain the traffic flow feature vector .

[0049] In this embodiment, the traffic flow feature vector in all the collected data is used as a data set, and K-means clustering is used to cluster out three cluster centers . The traffic flow states corresponding to the three cluster centers are: high density, medium density, and low density; subsequently, the traffic flow feature vector = { } is compared with to directly output the traffic flow state.

[0050] As Figure 5 shown, in this embodiment, the step 5 of multi-modal feature fusion for trajectory prediction specifically includes the following steps: Step 5.1. Use the Vector Net encoder to vectorize and encode the historical trajectory information and the high-precision map; Step 5.2. Use the LSTM encoder to encode the historical trajectory information to obtain the historical trajectory feature , and use the encoder based on point Net to encode the high-precision map information to obtain the map feature ; Step 5.3. Use the MLP (Multi-Layer Perceptron) to perform feature mapping on the dynamic feature vectors of N surrounding vehicles and the traffic flow feature vector = { } to obtain the dynamic features of the surrounding vehicles and traffic flow characteristics , the expression is:

[0051]

[0052] In the formula, represents a simple MLP, which consists of two linear transformation layers and a Relu activation function.

[0053] Step 5.4. Concatenate the historical trajectory features and the dynamic characteristics of the vehicle to obtain the feature . Concatenate the map feature with the traffic flow feature to obtain the feature , the expression is: , ) , ) In the formula, and are the features obtained after concatenation, is the historical trajectory feature, is the kinematic feature of the vehicle, is the map feature, is the traffic flow feature, is to concatenate the two feature vectors in sequence.

[0054] Step 5.5. Input the feature and the feature into the multi-head cross-attention module for interactive modeling to obtain the feature , the expression is:

[0055] In the formula, represents multi-head cross-attention, the number of multi-heads is 8, Q represents the query matrix, and the source is the feature , K is the key matrix, and the source is the feature , V is the value matrix, and the source is the feature ; Step 5.6. Use the feed-forward neural network to process the feature to obtain the feature , the expression is:

[0056] In the formula, is a feedforward neural network.

[0057] Step 5.7. Use the MLP to decode the features to decode the future trajectories of the N surrounding vehicles , the future speed

[0058]

[0059]

[0060] In the formula, }, }, represent the future trajectories of the N surrounding vehicles, = } represents the trajectory coordinates at the future time steps, = } represents the speed at the future time steps, = } represents the acceleration at the future time steps.

[0061] As Figure 6 shown, in this embodiment, the specific steps of the collision risk prediction in step 6 are as follows: Step 6.1. Calculate the distance from the N surrounding vehicles to the autonomous vehicle at the historical time step, and select the vehicle with the smallest distance from the autonomous vehicle as the target vehicle; among them, the distance from the surrounding vehicle i to the autonomous vehicle at the time step is expressed as:

[0062]

[0063] In the formula, represents the distance from the surrounding vehicle i to the autonomous vehicle at the time step, represents the x, y coordinates of the autonomous vehicle at the time step, represents the x, y coordinates of the surrounding vehicle i at the time step, , Represents the minimum distance between surrounding vehicles and the autonomous vehicle; Step 6.2. Based on the future speed of the target vehicle ={ }, acceleration , extract the maximum speed of the target vehicle , Maximum acceleration ; Step 6.3. Based on the future acceleration of the target vehicle Calculate the standard deviation of acceleration , the expression is:

[0064]

[0065] In the formula, represents the average acceleration, Represents the maximum future time step for prediction.

[0066] Step 6.4. Use the maximum speed of the target vehicle , Maximum acceleration and the standard deviation of acceleration Constructing a dynamic collision risk indicator , the expression is:

[0067] In the formula, , are the weight coefficients of velocity, acceleration, and fluctuation, which are set to 1.0, 1.2, and 1.0 respectively; Indicates the maximum speed limit of the road. is a safe acceleration, set , is the expected safety fluctuation value, set .

[0068] Step 6.5. Setting the dynamic fluctuation coefficient Corrected dynamic collision risk indicator , and the dynamic collision risk index after considering dynamic fluctuations is obtained , the expression is:

[0069] In the formula, Indicates the dynamic fluctuation coefficient, high dynamic fluctuation , mesodynamic fluctuations Low dynamic fluctuations ; Step 6.6. Set the traffic flow status label For the dynamic collision risk index after considering the dynamic fluctuations Make corrections to obtain the final dynamic collision risk index , and the expression is:

[0070] In the formula, represents the traffic flow status label. When it is in high density , when it is in medium density , when it is in low density , is the final dynamic risk index.

[0071] Step 6.7. Input the final dynamic risk index into the risk judgment module for judgment, and output the risk level , and the expression is: ； The technical solution provided by the present invention can effectively improve the recognition and prediction capabilities of the autonomous driving system in rare high-risk events, significantly improve the problem of insufficient prediction accuracy of the existing methods in long-tail scenarios, and has important safety value and broad engineering application prospects.

[0072] The present invention also provides an electronic device, including: one or more processors, a memory; wherein, the memory is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method for predicting the collision risk of an autonomous driving vehicle for the long-tail phenomenon.

[0073] The present invention also provides a computer-readable medium, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned method for predicting the collision risk of an autonomous driving vehicle for the long-tail phenomenon is implemented.

[0074] Those skilled in the art can understand that all or part of the functions of the above-mentioned methods / modules can be implemented in a hardware manner or in a computer program manner. When all or part of the functions in the above-mentioned embodiments are implemented in a computer program manner, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, magnetic disk, optical disk, hard disk, etc., and the above functions are implemented by a computer executing the program. For example, when the program is stored in the memory of the device and the program in the memory is executed by the processor, the above-mentioned all or part of the functions can be implemented.

[0075] In addition, when all or part of the functions in the above embodiments are implemented in the form of a computer program, the program can also be stored in a storage medium such as a server, another computer, a magnetic disk, an optical disk, a flash drive, or a mobile hard disk, and saved to the memory of a local device by downloading or copying, or the system of the local device can be updated. When the program in the memory is executed by a processor, all or part of the functions in the above embodiments can be realized.

[0076] The above uses specific examples to elaborate on the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the technical field to which the present invention pertains, based on the idea of the present invention, several simple deductions, deformations, or substitutions can also be made. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for predicting collision risk of an autonomous driving vehicle targeting the long-tail phenomenon, characterized in that: This method uses a multimodal enhanced trajectory dataset to train a collision risk prediction model, and then uses the trained collision risk prediction model to perform risk prediction. The specific risk prediction steps are: Step 1. Construct the dynamic feature vector of surrounding vehicles; Step 2. Output the dynamic fluctuation type according to the dynamic characteristic vectors of the surrounding vehicles; Step 3. Construct traffic flow feature vector; Step 4. Output the traffic flow state according to the traffic flow feature vector; Step 5. Trajectory prediction based on multimodal feature fusion: Encode the historical trajectory information of surrounding vehicles to obtain historical trajectory features , encode the high-precision map information to obtain map features ; Perform feature mapping on the dynamic feature vectors of surrounding vehicles and traffic flow feature vectors to obtain dynamic feature and traffic flow characteristics ; The historical trajectory characteristics and dynamic characteristics Splicing to get features , map features Traffic flow characteristics Splicing to get features ; The feature and Features Perform interactive modeling to obtain features ; Features Process and obtain features ; Features Decode the future trajectory, speed, and acceleration of surrounding vehicles; Step 6. Collision risk prediction: According to historical data, the vehicle with the shortest distance to the autonomous vehicle is selected as the target vehicle, the maximum speed and maximum acceleration of the target vehicle are extracted, and the standard deviation of acceleration is calculated; a dynamic collision risk index is constructed based on the maximum speed, maximum acceleration and standard deviation of acceleration of the target vehicle; finally, the dynamic fluctuation coefficient is determined according to the dynamic fluctuation type, the traffic flow state label is determined according to the traffic flow state, and the dynamic collision risk index is corrected using the dynamic fluctuation coefficient and the traffic flow state label.

2. The method for predicting collision risk of an autonomous driving vehicle for the long tail phenomenon according to claim 1, characterized in that: The steps for constructing the multimodal enhanced trajectory dataset are as follows: Step A. Data collection and feature construction: step A1. Filter trajectory data and construct tail trajectory data samples; Step A2: normalizing the selected tail trajectory data; Step A3. Trajectory feature extraction; Extracting multi-dimensional features of target vehicles from standardized tail trajectory data ; Wherein, P represents the position sequence of the target vehicle, V represents the velocity sequence of the target vehicle, A represents the acceleration sequence of the target vehicle, and K represents the curvature sequence of the target vehicle; Step B. Data enhancement based on diffusion model: Step B1. Based on the constructed multi-dimensional features of the target vehicle , adding Gaussian noise at each time step , get the features ; Step B2. Use the Transformer neural network to predict the Gaussian noise added at each time step , then for the features Each time step feature Denoise and get features ; Step B3. De-noised features Input into the multi-layer perceptron to generate trajectory data; Step B4. Sampling different Gaussian noises multiple times , repeat the process from step B2 to step B3 to generate multiple different trajectory samples, thereby constructing a multimodal enhanced trajectory dataset.

3. The method for predicting collision risk of an autonomous driving vehicle for the long tail phenomenon according to claim 1, characterized in that: Step 1 Calculate the average speed based on the historical data of surrounding vehicles i , mean acceleration , speed standard deviation , Peak speed , velocity skewness , speed variation coefficient , construct the dynamic characteristic vector of the surrounding vehicle i ; Step 2: The dynamic feature vector of the surrounding vehicle i The cluster centers representing high dynamic fluctuations, medium dynamic fluctuations, and low dynamic fluctuations Compare and output the dynamic fluctuation type of the surrounding vehicle i.

4. The method for predicting collision risk of an autonomous driving vehicle for the long tail phenomenon according to claim 1, characterized in that: Step 3: Calculate the traffic density based on the sensing range radius R of the sensor on the autonomous vehicle and the number of vehicles N within the sensing range. , Traffic flow , the average speed of surrounding vehicles , construct traffic flow feature vector ={ }; Step 4: Transform the traffic flow feature vector around the autonomous vehicle And the cluster centers representing high density, medium density, and low density Compare and output the traffic flow status.

5. The method for predicting collision risk of an autonomous driving vehicle for the long tail phenomenon according to claim 1, characterized in that: In step 5, the LSTM encoder is used to encode the historical trajectory information to obtain the historical trajectory features , use the point Net-based encoder to encode the high-precision map information to obtain map features ; Use a multi-layer perceptron to perform feature mapping on the dynamic feature vectors of surrounding vehicles and traffic flow feature vectors to obtain the dynamic features of surrounding vehicles and traffic flow characteristics .

6. The method for predicting collision risk of an autonomous driving vehicle for the long tail phenomenon according to claim 1, characterized in that: In step 5, the features and Features Input to the multi-head cross attention module for interactive modeling to obtain features ; Use feedforward neural network to train features Process and obtain features ; Use multi-layer perceptron to analyze features to decode.

7. The method for predicting collision risk of an autonomous driving vehicle for the long tail phenomenon according to claim 1, characterized in that: The specific steps of the collision risk prediction in step 6 are as follows: Step 6.

1. Calculate the The distances from the N surrounding vehicles to the autonomous vehicle at any given moment, and the vehicle with the shortest distance from the autonomous vehicle is selected as the target vehicle; Step 6.

2. Extract the maximum speed of the target vehicle based on the future speed and acceleration of the target vehicle , maximum acceleration ; Step 6.

3. Calculate the acceleration standard deviation based on the future acceleration of the target vehicle; Step 6.

4. Use the maximum speed of the target vehicle , maximum acceleration and the standard deviation of acceleration Constructing a dynamic collision risk indicator , the expression is: ； In the formula, , is the weight coefficient of velocity, acceleration and fluctuation, Indicates the maximum speed limit of the road. For safe acceleration, is the expected safety fluctuation value; Step 6.

5. Setting the kinetic fluctuation coefficient Corrected dynamic collision risk indicator , and the dynamic collision risk index after considering dynamic fluctuations is obtained , the expression is: ； Step 6.

6. Set traffic flow status label Dynamic collision risk index after considering dynamic fluctuations Make corrections to obtain the final dynamic collision risk index , the expression is: ； Step 6.

7. The final dynamic risk indicator Input to the risk judgment module for interpretation and output of risk level .

8. The method for predicting collision risk of an autonomous driving vehicle for the long tail phenomenon according to claim 3, characterized in that: The peak speed in step 1 , velocity skewness , speed variation coefficient The expression is shown as: ； ； ； in, represents the velocity at the f-th time step, is the historical trajectory time step.

9. The method for predicting collision risk of an autonomous driving vehicle for the long tail phenomenon according to claim 7, characterized in that: In step 6.4 , Set to 1.0, 1.2, 1.0 respectively; Set as , Set as ; Step 6.5 Medium and high dynamic fluctuations , mesodynamic fluctuations Low dynamic fluctuations ; Step 6.6: Medium to high density , medium density , at low density .

Images