Vehicle track prediction method and system

Abstract

The invention provides a vehicle trajectory prediction method and system, and the method comprises the steps: firstly, obtaining a trajectory prediction sample data set; then, data preprocessing is carried out, and a preprocessed trajectory prediction sample data set is divided into a training data set and a test data set; secondly, constructing an end-to-end trajectory prediction model; and training the end-to-end trajectory prediction model according to the training data set, and determining a training loss rate. And testing the trained end-to-end trajectory prediction model according to the test data set, and determining a test loss rate. And finally, under the condition that the training loss rate and the test loss rate are in a preset loss rate range, inputting to-be-predicted trajectory data into the trained end-to-end trajectory prediction model, outputting a steering wheel rotation angle prediction result, and determining a vehicle trajectory prediction result. Therefore, the architecture of the trajectory prediction system is simplified, error propagation is reduced, and the robustness and adaptability of the end-to-end trajectory prediction model are enhanced, so that the accuracy of vehicle trajectory prediction is improved.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a vehicle trajectory prediction method and system. Background Art

[0002] With the rapid development of artificial intelligence, autonomous driving technology is being widely used. Vehicle trajectory prediction is a crucial component of autonomous driving. It predicts the vehicle's future trajectory based on its external driving environment, ensuring that the autonomous driving system can make correct driving decisions in advance, reducing the likelihood of accidents and improving driving safety.

[0003] Typically, vehicle trajectory prediction methods rely on modular trajectory prediction systems, breaking down the task into multiple modules, including perception, localization, planning, and control. This leads to errors, complex system integration, and insufficient robustness and adaptability, resulting in low trajectory prediction accuracy. Summary of the Invention

[0004] The technical problem to be solved by the present invention is the inability to accurately predict vehicle trajectories.

[0005] To solve the above technical problems, the present invention provides a vehicle trajectory prediction method and system, which specifically adopts the following technical solutions: In a first aspect, the present invention provides a vehicle trajectory prediction method, comprising: first, obtaining a trajectory prediction sample dataset, the trajectory prediction sample dataset comprising: sample road condition images from multiple vehicle perspectives, as well as sample vehicle speeds and sample steering wheel angles corresponding to the road condition sample images. Then, preprocessing the trajectory prediction sample dataset to obtain a preprocessed trajectory prediction sample dataset, and dividing the preprocessed trajectory prediction sample dataset into a training dataset and a test dataset according to a preset partitioning ratio. Next, constructing an end-to-end trajectory prediction model, the end-to-end trajectory prediction model being used to predict the vehicle's steering wheel angle. Next, training the end-to-end trajectory prediction model based on the training dataset, determining a training loss rate, and obtaining a trained end-to-end trajectory prediction model. Testing the trained end-to-end trajectory prediction model based on the test dataset, and determining a test loss rate. Finally, if the training loss rate and the test loss rate are within a preset loss rate range, inputting the trajectory data to be predicted into the trained end-to-end trajectory prediction model, outputting a steering wheel angle prediction result, and determining a vehicle trajectory prediction result based on the steering wheel angle prediction result.

[0006] This method employs an end-to-end vehicle trajectory prediction method, using an end-to-end trajectory prediction model to perform deep learning and training on a trajectory prediction sample dataset. Comprehensive data augmentation techniques improve the generalization and robustness of the end-to-end trajectory prediction model, and a custom self-attention layer within the end-to-end trajectory prediction model enhances feature representation. Furthermore, the trained end-to-end trajectory prediction model can derive steering wheel angle predictions based on the trajectory data to be predicted collected from the sensor. Finally, the predicted vehicle trajectory is determined based on the steering wheel angle prediction results. This vehicle trajectory prediction method simplifies the trajectory prediction system architecture, reduces error propagation, and enhances the robustness and adaptability of the end-to-end trajectory prediction model, thereby improving the accuracy of vehicle trajectory prediction.

[0007] In combination with the first aspect, in an optional implementation, the above-mentioned end-to-end trajectory prediction model is a convolutional neural network structure, and the end-to-end trajectory prediction model includes: multiple convolutional layers, maximum pooling layers, ReLU activation functions, Dropout layers, self-attention layers and fully connected layers.

[0008] In conjunction with the first aspect, in an optional implementation, training the end-to-end trajectory prediction model based on the training dataset, determining the training loss rate, and obtaining the trained end-to-end trajectory prediction model includes: inputting the training dataset into the end-to-end trajectory prediction model to obtain a training prediction result; iteratively optimizing model parameters of the end-to-end trajectory prediction model based on the training prediction result to obtain the training loss rate and the trained end-to-end trajectory prediction model.

[0009] In combination with the first aspect, in an optional implementation, the above-mentioned multiple vehicle perspectives include: a vehicle left front perspective, a vehicle front perspective, and a vehicle right front perspective.

[0010] In combination with the first aspect, in an optional implementation, the batch size of the end-to-end trajectory prediction model trained according to the training dataset is 32, the number of training rounds is 500, and the learning rate is 0.001.

[0011] In combination with the first aspect, in an optional implementation, the training of the end-to-end trajectory prediction model based on the training dataset adopts an early termination strategy, and the tolerance for early termination is 300 cycles.

[0012] In combination with the first aspect, in an optional implementation, the discard ratio of the road condition sample images with a sample steering wheel angle of zero in the training of the end-to-end trajectory prediction model based on the training dataset is 95%.

[0013] In combination with the first aspect, in an optional implementation, the range of the sample vehicle speed is: 25km / h-30km / h.

[0014] In a second aspect, the present invention provides a vehicle trajectory prediction system, comprising: an acquisition module, a data preprocessing module, a model construction module, a model training module, a model testing module, and a prediction module. The acquisition module is configured to acquire a trajectory prediction sample dataset, which includes sample road condition images from multiple vehicle perspectives, as well as sample vehicle speeds and sample steering wheel angles corresponding to the acquisition of the road condition sample images. The data preprocessing module is configured to perform data preprocessing on the trajectory prediction sample dataset to obtain a preprocessed trajectory prediction sample dataset, and then divide the preprocessed trajectory prediction sample dataset into a training dataset and a test dataset according to a preset division ratio. The model construction module is configured to construct an end-to-end trajectory prediction model, which is used to predict the vehicle's steering wheel angle. The model training module is configured to train the end-to-end trajectory prediction model based on the training dataset to obtain a trained end-to-end trajectory prediction model. The model testing module is configured to test the trained end-to-end trajectory prediction model based on the test dataset and determine the test loss rate. The prediction module is used to input the trajectory data to be predicted into the trained end-to-end trajectory prediction model when the test loss rate meets the preset accuracy condition, output the steering wheel angle prediction result, and determine the vehicle trajectory prediction result based on the steering wheel angle prediction result.

[0015] In a third aspect, an electronic device is provided, comprising: a memory, one or more processors; the memory and the processor are coupled; wherein computer program code is stored in the memory, and the computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device executes any one of the methods described in the first aspect above.

[0016] It can be understood that the beneficial effects that can be achieved by the vehicle trajectory prediction system provided by the second aspect and the electronic device provided by the third aspect can refer to the beneficial effects in the first aspect and any possible design method thereof, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a process flow of a vehicle trajectory prediction method provided by an embodiment of the present invention; Figure 2 A schematic diagram of the training loss rate and test loss rate of the end-to-end trajectory prediction model training provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a vehicle trajectory prediction system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following embodiments are described in detail, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numbers in different figures represent identical or similar elements unless otherwise indicated. The embodiments described in the following embodiments are not intended to represent all possible implementations consistent with the present application. They are merely examples of systems and methods consistent with certain aspects of the present application, as detailed in the claims.

[0019] With the rapid development of artificial intelligence, autonomous driving technology is being widely used. Vehicle trajectory prediction is a crucial component of autonomous driving. It predicts the vehicle's future trajectory based on its external driving environment, ensuring that the autonomous driving system can make correct driving decisions in advance, reducing the likelihood of accidents and improving driving safety.

[0020] Typically, vehicle trajectory prediction methods rely on modular trajectory prediction systems, breaking down the task into multiple modules, including perception, localization, planning, and control. This leads to errors, complex system integration, and insufficient robustness and adaptability, resulting in low trajectory prediction accuracy.

[0021] In order to solve the above problems, the embodiments of the present application provide a vehicle trajectory prediction method and system. Specifically, the method adopts an end-to-end vehicle trajectory prediction method, and performs deep learning and training on the trajectory prediction sample data set through the end-to-end trajectory prediction model. Furthermore, the trained end-to-end trajectory prediction model can obtain the steering wheel angle prediction result based on the trajectory data to be predicted collected from the sensor. Finally, the predicted vehicle trajectory is determined based on the steering wheel angle prediction result. In this way, the vehicle trajectory prediction method simplifies the trajectory prediction system architecture, reduces error propagation, enhances the robustness and adaptability of the end-to-end trajectory prediction model, and thus improves the accuracy of vehicle trajectory prediction.

[0022] The vehicle trajectory prediction method and system provided in the embodiments of the present application can be applied to vehicle trajectory prediction on complex and changeable roads (e.g., rural roads), which usually have features such as unpaved roads, narrow road widths, sharp turns, slopes, and unclear intersections.

[0023] The following describes the solution provided by the embodiments of the present application in conjunction with the accompanying drawings.

[0024] For details, see Figure 1 , which is a flow chart of a vehicle trajectory prediction method provided by an embodiment of the present invention, such as Figure 1 As shown, the vehicle trajectory prediction method provided by the present invention includes the following steps S101-S106: S101: Obtain a trajectory prediction sample dataset.

[0025] The trajectory prediction sample data set includes: a plurality of road condition sample images from different vehicle perspectives, and corresponding sample vehicle speeds and sample steering wheel rotation angles when the road condition sample images are collected. Specifically, the road condition sample images are used to represent different road conditions outside the vehicle under different driving conditions (e.g., straight driving, left turn, right turn, etc.). The sample vehicle speed is used to represent the driving speed of the vehicle when the road condition sample image is collected. The sample steering wheel rotation angle is used to represent the rotation angle and rotation direction of the steering wheel of the vehicle when the road condition sample image is collected. For example, the sample steering wheel rotation angle can be +20 degrees (i.e., indicating a clockwise rotation of 20 degrees), and the sample steering wheel rotation angle can also be -15 degrees (i.e., indicating a counterclockwise rotation of 15 degrees).

[0026] In some embodiments, the plurality of vehicle perspectives includes: a vehicle left front perspective, a vehicle front perspective, and a vehicle right front perspective. Specifically, other vehicle perspectives can also be selected according to the actual application requirements.

[0027] In an implementation manner, the trajectory prediction sample data set is obtained by a vehicle simulator.

[0028] In some embodiments, the sample vehicle speed can be in the range of 25 km / h-30 km / h, i.e., when the trajectory prediction sample data set is obtained, the sample vehicle (e.g., a vehicle simulator) collects the road condition sample image at a speed of 25 km / h-30 km / h. In this way, keeping the vehicle speed at 25 km / h-30 km / h can provide a relatively stable driving environment and collection environment, which can improve the prediction accuracy of the constructed end-to-end trajectory prediction model.

[0029] S102, data preprocessing is performed on the trajectory prediction sample data set to obtain a preprocessed trajectory prediction sample data set, and the preprocessed trajectory prediction sample data set is divided into a training data set and a test data set according to a preset division ratio.

[0030] Specifically, the obtained trajectory prediction sample data set is preprocessed. In order to facilitate the training and optimization of the end-to-end trajectory prediction model, the preprocessed trajectory prediction sample data set can be divided into a training data set and a test data set.

[0031] In some embodiments, the data preprocessing of the trajectory prediction sample data set includes data cleaning, data enhancement, and data labeling. Specifically, the cleaning process includes removing invalid data or samples with obvious errors in the trajectory prediction sample data set. Data enhancement can expand the data volume by adjusting the brightness and contrast of the road condition sample images in the trajectory prediction sample data set. In the labeling process, the actual trajectory information of the vehicle, such as the steering angle and the speed, is added to each road condition sample image.

[0032] In some embodiments, the preset division ratio may be 7:3 for training dataset:test dataset. In this way, the end-to-end trajectory prediction model can be effectively trained with a relatively large training dataset, thereby improving the accuracy of the end-to-end trajectory prediction model.

[0033] S103: Build an end-to-end trajectory prediction model, where the end-to-end trajectory prediction model is used to predict the steering wheel angle of the vehicle.

[0034] In some embodiments, the end-to-end trajectory prediction model is a convolutional neural network (CNN) structure, and the end-to-end trajectory prediction model includes: multiple convolutional layers, maximum pooling layers, ReLU activation functions, dropout layers, self-attention layers, and fully connected layers.

[0035] Specifically, the end-to-end trajectory prediction model may include: multiple convolutional layers and maximum pooling layers, combined with ReLU activation functions and Dropout (regularization) layers to extract image features from road images and prevent overfitting. The end-to-end trajectory prediction model also adds a self-attention layer with customized hyperparameters and network structure to enhance the expressiveness of features. The fully connected layer further processes the features and ultimately outputs a single linearly activated value as the steering wheel angle prediction result.

[0036] S104: Train the end-to-end trajectory prediction model according to the training data set, determine the training loss rate, and obtain the trained end-to-end trajectory prediction model.

[0037] Furthermore, the end-to-end trajectory prediction model can be trained based on the training dataset. The model parameters of the end-to-end trajectory prediction model can be adjusted through the input training dataset data and the output training prediction results, so that the end-to-end trajectory prediction model can learn the characteristics and rules of the training dataset.

[0038] In some embodiments, S104 may specifically include: First, the training dataset is input into the end-to-end trajectory prediction model to obtain the training prediction results.

[0039] Then, the model parameters of the end-to-end trajectory prediction model are iteratively optimized according to the training prediction results to obtain the training loss rate and the trained end-to-end trajectory prediction model.

[0040] In some embodiments, when the end-to-end trajectory prediction model is trained based on the training dataset, the size of the road condition sample images input into the end-to-end trajectory prediction model in the training dataset is 128 (pixels)×128 (pixels)×3 (channels).

[0041] In some embodiments, the end-to-end trajectory prediction model is trained based on the training dataset with a batch size of 32, a number of training rounds of 500, and a learning rate of 0.001.

[0042] In some embodiments, when training the end-to-end trajectory prediction model based on the training dataset, the Adam optimizer can be used to compile the model, and an early termination strategy is adopted during the training optimization process, with a tolerance for early termination of 300 cycles.

[0043] In some embodiments, the end-to-end trajectory prediction model is trained based on the training dataset, and during the data enhancement process, the discard ratio of road condition sample images with a sample steering wheel angle of zero is 95%.

[0044] S105: Test the trained end-to-end trajectory prediction model based on the test data set and determine the test loss rate.

[0045] Next, the trained end-to-end trajectory prediction model can be tested on the test dataset and the test loss rate can be determined. This is used to evaluate the performance of the trained end-to-end trajectory prediction model and to examine its generalization ability to unseen data (such as the test dataset data). This is used to measure whether the trained end-to-end trajectory prediction model can accurately predict new data.

[0046] S106. When the training loss rate and the test loss rate are within a preset loss rate range, the trajectory data to be predicted is input into the trained end-to-end trajectory prediction model, a steering wheel angle prediction result is output, and a vehicle trajectory prediction result is determined based on the steering wheel angle prediction result.

[0047] Specifically, the preset loss rate ranges corresponding to the training loss rate and the test loss rate can be the same or different. If the training loss rate and the test loss rate are within the preset loss rate range, it can be determined that the trained end-to-end trajectory prediction model can meet the application requirements.

[0048] For example, the preset loss rate range can be preset according to different application scenarios and application requirements. For example, the preset loss rate range can be set according to different road conditions. The preset loss rate range can be, for example, 0.03-0.05.

[0049] Figure 2 Schematic diagram of the training loss rate and test loss rate of the end-to-end trajectory prediction model training provided by the embodiment of the present invention. Figure 2As shown, curve 201 represents the training loss rate, and curve 202 represents the test loss rate. As the number of training times increases, the training and test loss rates gradually stabilize. When the preset loss rate range is 0.03-0.07, the training and test loss rates are within the preset loss rate range, indicating that the trained end-to-end trajectory prediction model can meet application requirements.

[0050] Furthermore, when the test loss rate meets the preset accuracy condition, the trajectory data to be predicted is input into the trained end-to-end trajectory prediction model, and the steering wheel angle prediction result is output.

[0051] Finally, the vehicle trajectory prediction result can be determined based on the steering wheel angle prediction result and the vehicle speed to determine the distance and direction of the vehicle.

[0052] The trajectory data to be predicted may include real-time images of road conditions from the same perspective as the multiple vehicle perspectives in the trajectory prediction sample dataset. For example, the trajectory data to be predicted may include real-time images of road conditions from the vehicle's left front perspective, real-time images of road conditions from the vehicle's front perspective, and real-time images of road conditions from the vehicle's right front perspective.

[0053] The vehicle trajectory prediction method provided by the embodiment of the present invention adopts an end-to-end vehicle trajectory prediction method, and performs deep learning and training on the trajectory prediction sample data set through the end-to-end trajectory prediction model. Comprehensive data enhancement technology improves the generalization ability and robustness of the end-to-end trajectory prediction model, and the customized self-attention layer in the end-to-end trajectory prediction model enhances the feature expression ability. Furthermore, the trained end-to-end trajectory prediction model can obtain the steering wheel angle prediction result based on the trajectory data to be predicted collected from the sensor. Finally, the predicted vehicle trajectory is determined based on the steering wheel angle prediction result. In this way, the vehicle trajectory prediction method simplifies the trajectory prediction system architecture, reduces error propagation, enhances the robustness and adaptability of the end-to-end trajectory prediction model, and thus improves the accuracy of vehicle trajectory prediction.

[0054] The embodiment of the present invention further provides a vehicle trajectory prediction system, specifically, Figure 3 A schematic diagram of the structure of a vehicle trajectory prediction system provided by an embodiment of the present invention is shown in FIG. Figure 3 As shown, the vehicle trajectory prediction system 300 includes: an acquisition module 301, a data preprocessing module 302, a model construction module 303, a model training module 304, a model testing module 305 and a prediction module 306.

[0055] The acquisition module 301 may be used to acquire a trajectory prediction sample data set, which includes: sample road condition images from multiple vehicle perspectives, and sample vehicle speeds and sample steering wheel angles corresponding to the acquisition of the sample road condition images.

[0056] The data preprocessing module 302 may be used to perform data preprocessing on the trajectory prediction sample dataset to obtain a preprocessed trajectory prediction sample dataset, and divide the preprocessed trajectory prediction sample dataset into a training dataset and a test dataset according to a preset division ratio.

[0057] The model building module 303 can be used to build an end-to-end trajectory prediction model, which is used to predict the steering wheel angle of the vehicle.

[0058] The model training module 304 may be configured to train the end-to-end trajectory prediction model according to the training data set, determine the training loss rate, and obtain the trained end-to-end trajectory prediction model.

[0059] The model testing module 305 may be used to test the trained end-to-end trajectory prediction model based on a test dataset and determine a test loss rate.

[0060] The prediction module 306 can be used to input the trajectory data to be predicted into the trained end-to-end trajectory prediction model when the training loss rate and the test loss rate are within a preset loss rate range, output a steering wheel angle prediction result, and determine the vehicle trajectory prediction result based on the steering wheel angle prediction result.

[0061] The vehicle trajectory prediction system provided by the embodiments of the present invention can perform deep learning and training on a trajectory prediction sample dataset using an end-to-end trajectory prediction model. The trained end-to-end trajectory prediction model can then derive a steering wheel angle prediction result based on the trajectory data to be predicted, collected from sensors. Finally, the predicted vehicle trajectory is determined based on the steering wheel angle prediction result. This vehicle trajectory prediction method simplifies the trajectory prediction system architecture, reduces error propagation, and enhances the robustness and adaptability of the end-to-end trajectory prediction model, thereby improving the accuracy of vehicle trajectory prediction.

[0062] An embodiment of the present invention further provides an electronic device, which may include: a display screen, a memory, and one or more processors. The display screen, memory, and processor are coupled. The memory is configured to store computer program code, which includes computer instructions. When the processor executes the computer instructions, the electronic device may perform the various methods or steps described in the above-described embodiment of the vehicle trajectory prediction method. Of course, the electronic device includes, but is not limited to, the above-described display screen, memory, and one or more processors.

[0063] An embodiment of the present invention further provides a computer-readable storage medium for storing computer instructions for executing the above-mentioned vehicle trajectory prediction method.

[0064] Through the description of the above implementation methods, technical personnel in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0065] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. In the description of the present invention, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0066] In the present invention, unless otherwise specified or limited, the terms "installed," "connected," "connect," "fixed," etc. should be understood in a broad sense. For example, they can refer to fixed connection, detachable connection, or integration; mechanical connection, electrical connection; direct connection, or indirect connection through an intermediate medium; internal communication between two components, or interaction between two components, unless otherwise specified. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0067] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0068] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the overall concept of this application and do not constitute a limitation on the scope of protection of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without expending creative work shall fall within the scope of protection of this application.

Claims

1. A vehicle trajectory prediction method, characterized in that: include: Acquire a trajectory prediction sample data set, the trajectory prediction sample data set comprising: road condition sample images from multiple vehicle perspectives, and sample vehicle speeds and sample steering wheel angles corresponding to when the road condition sample images are collected; Performing data preprocessing on the trajectory prediction sample data set to obtain a preprocessed trajectory prediction sample data set, and dividing the preprocessed trajectory prediction sample data set into: a training data set and a test data set according to a preset division ratio; Building an end-to-end trajectory prediction model, wherein the end-to-end trajectory prediction model is used to predict a steering wheel angle of a vehicle; Training the end-to-end trajectory prediction model according to the training data set, determining a training loss rate, and obtaining a trained end-to-end trajectory prediction model; Testing the trained end-to-end trajectory prediction model based on the test dataset and determining a test loss rate; When the training loss rate and the test loss rate are within a preset loss rate range, the trajectory data to be predicted is input into the trained end-to-end trajectory prediction model, a steering wheel angle prediction result is output, and a vehicle trajectory prediction result is determined based on the steering wheel angle prediction result.

2. The method according to claim 1, characterized in that The end-to-end trajectory prediction model is a convolutional neural network structure, which includes: multiple convolutional layers, maximum pooling layers, ReLU activation functions, Dropout layers, self-attention layers and fully connected layers.

3. The method according to claim 1 or 2, characterized in that The step of training the end-to-end trajectory prediction model according to the training data set, determining a training loss rate, and obtaining a trained end-to-end trajectory prediction model includes: Inputting the training data set into the end-to-end trajectory prediction model to obtain a training prediction result; Iteratively optimize model parameters of the end-to-end trajectory prediction model according to the training prediction results to obtain a training loss rate and a trained end-to-end trajectory prediction model.

4. The method according to claim 1, wherein The multiple vehicle perspectives include: a vehicle left front perspective, a vehicle front perspective, and a vehicle right front perspective.

5. The method according to claim 1, wherein The batch size for training the end-to-end trajectory prediction model based on the training dataset is 32, the number of training rounds is 500, and the learning rate is 0.

001.

6. The method according to claim 1, characterized in that The end-to-end trajectory prediction model is trained according to the training data set using an early termination strategy, and the tolerance for early termination is 300 cycles.

7. The method according to claim 1, characterized in that The discard ratio of the road condition sample images in which the sample steering wheel angle is zero in the training of the end-to-end trajectory prediction model according to the training data set is 95%.

8. The method according to claim 1, characterized in that The sample vehicle speed range is: 25km / h-30km / h.

9. A vehicle trajectory prediction system, characterized in that: include: Acquisition module, data preprocessing module, model building module, model training module, model testing module and prediction module; among them, The acquisition module is used to acquire a trajectory prediction sample data set, wherein the trajectory prediction sample data set includes: road condition sample images from multiple vehicle perspectives, and sample vehicle speeds and sample steering wheel angles corresponding to when the road condition sample images are collected; The data preprocessing module is used to perform data preprocessing on the trajectory prediction sample data set to obtain a preprocessed trajectory prediction sample data set, and divide the preprocessed trajectory prediction sample data set into: a training data set and a test data set according to a preset division ratio; The model building module is used to build an end-to-end trajectory prediction model, and the end-to-end trajectory prediction model is used to predict the steering wheel angle of the vehicle; The model training module is used to train the end-to-end trajectory prediction model according to the training data set, determine the training loss rate, and obtain a trained end-to-end trajectory prediction model; The model testing module is used to test the trained end-to-end trajectory prediction model based on the test dataset and determine a test loss rate; The prediction module is configured to input the trajectory data to be predicted into the trained end-to-end trajectory prediction model when the training loss rate and the test loss rate are within a preset loss rate range, output a steering wheel angle prediction result, and determine a vehicle trajectory prediction result based on the steering wheel angle prediction result.

10. An electronic device, characterized in that: include: A memory, one or more processors; the memory is coupled to the processor; wherein the memory stores computer program code, the computer program code includes computer instructions, and when the computer instructions are executed by the processor, the electronic device executes the method according to any one of claims 1 to 8.

Images