Steering wheel angle calculation model training method, display method, system and terminal

Abstract

The invention provides a steering wheel angle calculation model training method, a display method, a system and a terminal. The steering wheel angle calculation model training method comprises the steps that an angle calculation model and a corresponding reward function are initialized, a steering wheel prediction angle is acquired based on the angle calculation model and historical image data, and the reward function is updated based on the steering wheel prediction angle and a pre-stored steering wheel actual angle; and when values corresponding to the reward function belong to a preset range after the reward function is iterated for a preset time, judging that the angle calculation model training is completed. The angle calculation model obtained through training can keep high accuracy in different environments and conditions, the model can effectively deal with complex road conditions, and precision reduction caused by environment changes is avoided.

Description

Technical Field

[0001] This application relates to the field of vehicle auxiliary technologies, and particularly to a method for training a steering wheel angle calculation model, a display method, a system, and a terminal. Background Art

[0002] Autonomous driving technology is a technology that enables a vehicle to drive autonomously through advanced sensors, computer vision, artificial intelligence, and control systems. The goal of this technology is to enable the vehicle to drive safely and efficiently under various traffic and road conditions without human intervention.

[0003] One of the cores of autonomous driving technology is to accurately calculate the rotation angle of the vehicle's steering wheel to achieve precise vehicle control. Currently, most methods rely on sensor data, but these methods have limited performance in complex environments. Summary of the Invention

[0004] To solve the above technical problems, this application provides a method for training a steering wheel angle calculation model, a display method, a system, and a terminal that can improve the accuracy of autonomous driving.

[0005] Specifically, this application provides a method for training a steering wheel angle calculation model, including:

[0006] Initializing an angle calculation model and a corresponding reward function, and obtaining a predicted steering wheel angle based on the angle calculation model and historical image data, so as to update the reward function based on the predicted steering wheel angle and the actually stored steering wheel angle. When the values of the reward function all belong to a preset range after a preset number of iterations, it is determined that the training of the angle calculation model is completed.

[0007] In the above technical solution, by introducing an angle calculation model and a reward function, the rotation angle of the steering wheel can be predicted more accurately. The model is continuously optimized during the training process to ensure that the error between the predicted steering wheel angle and the actual steering wheel angle is minimized. This method iteratively trains the reward function to maintain high accuracy in different environments and conditions. Even in complex road conditions, the model can effectively respond and avoid a decrease in accuracy caused by environmental changes.

[0008] Further, before initializing the angle calculation model, it includes:

[0009] Pre-collecting a number of historical image data with the steering wheel at different rotation angles, and storing the rotation angle corresponding to each historical image data as the actual steering wheel angle.

[0010] In the above technical solution, by collecting historical image data of the steering wheel at different rotation angles, the diversity of training data is ensured. This diversity helps the model to make accurate predictions at different rotation angles and avoids the overfitting problem that may be caused by a single-angle data set.

[0011] Further, the initialization reward function includes:

[0012] Initialize the predicted steering wheel angle and the actual steering wheel angle, and initialize the reward function based on the difference between the initialized predicted steering wheel angle and the actual steering wheel angle.

[0013] In the above technical solution, by calculating the difference between the initialized predicted steering wheel angle and the actual steering wheel angle, the error situation of the model in the initial state can be directly reflected. This initialization method based on error enables the reward function to be optimized for the actual performance of the model from the very beginning, avoiding the problem of the disconnection between the reward function and the model performance.

[0014] Further, the initialization angle calculation model includes:

[0015] Initialize the policy function, and initialize the angle calculation model based on the initialized policy function and the reward function.

[0016] In the above technical solution, by combining the policy function and the reward function, the angle calculation model adopts the framework of reinforcement learning. This framework allows the model to learn the optimal policy through interaction with the environment, rather than just learning based on static data, so as to better adapt to the dynamic environment.

[0017] Different from traditional supervised learning methods, reinforcement learning focuses on long-term cumulative rewards. Therefore, the model will consider the long-term impact of actions during the training process, which is particularly important for scenarios such as autonomous driving that require considering multiple future steps of operations.

[0018] Further, the angle calculation model at least adopts a convolutional neural network model, and the convolutional neural network model at least includes a convolutional layer and a fully connected layer.

[0019] The obtaining of the predicted steering wheel angle includes:

[0020] Input the historical image data into the initialized or updated angle calculation model, extract the image features corresponding to the historical image data based on the convolutional layer, and map the image features to the predicted steering wheel angle based on the fully connected layer.

[0021] In the above technical solution, the unique structure of the convolutional layer can efficiently extract local features (such as edges, textures, shapes, etc.) and global features in the image. Through multiple convolutional operations, the model can gradually extract high-level features (contours, structures) from low-level features (edges, lines), so as to more comprehensively understand the features of the steering wheel image. This efficient feature extraction ability enables the model to more accurately identify the rotation angle of the steering wheel, especially performing well in complex image backgrounds; the fully connected layer can map the rich features extracted by the convolutional layer to the specific predicted angle of the steering wheel through a non-linear activation function (such as ReLU) and weight adjustment. The powerful mapping ability of the fully connected layer enables the model to handle complex angle prediction tasks and continuously approximate the real rotation angle of the steering wheel by optimizing the weight parameters.

[0022] Further, it also includes: initializing the accumulated reward based on the initialized reward function.

[0023] In the above technical solution, by initializing the accumulated reward based on the initialized reward function, it can directly reflect the performance level of the model in the initial state, which helps to evaluate the initial performance of the model and provide a benchmark for subsequent training and optimization.

[0024] Further, after updating the reward function, it includes:

[0025] Updating the accumulated reward based on the updated reward function.

[0026] Calculating the policy gradient based on the updated reward function, the current policy function and the accumulated reward, and updating the policy function according to the policy gradient.

[0027] And, updating the angle calculation model based on the updated reward function and policy function.

[0028] In the above technical solution, by calculating the policy gradient based on the updated reward function, the current policy function and the accumulated reward, and updating the policy function according to the policy gradient, the model can continuously optimize its policy during the training process. This gradient-based update method can make the model gradually tend to the optimal policy; dynamically updating the reward function and the policy function can reduce the risk of model overfitting. Overfitting usually occurs when the model over-adapts to the training data and loses its generalization ability. Through continuous optimization and update, the model can better adapt to new data and new environments.

[0029] Further, based on the same concept, the present application also provides a method for displaying the rotation angle of a steering wheel, including: obtaining a steering wheel image in real time, and inputting the steering wheel image into the angle calculation model to output and display the steering wheel rotation angle information; wherein, the angle calculation model is trained by the steering wheel angle calculation model training method.

[0030] In the above technical solution, the predicted steering wheel rotation angle is displayed, providing immediate feedback and monitoring, thereby improving the accuracy of the steering wheel control of the autonomous driving system.

[0031] Furthermore, based on the same concept, the present application also provides a steering wheel rotation angle display system, including:

[0032] An acquisition module, configured to acquire a steering wheel image in real time and send the steering wheel image to a calculation module.

[0033] The calculation module is deployed with the angle calculation model to calculate the steering wheel rotation angle information based on the steering wheel image according to the angle calculation model, and output the steering wheel rotation angle information to a display module.

[0034] And the display module, configured to display the steering wheel rotation angle information.

[0035] In the above technical solution, the acquisition module can acquire the steering wheel image in real time, ensuring that the system can capture the dynamic changes of the steering wheel in a timely manner; the calculation module is deployed with an efficient angle calculation model (such as a convolutional neural network), which can complete image feature extraction and angle calculation in a short time, ensuring the real-time output of the rotation angle information; the display module can display the calculated steering wheel rotation angle information in real time, providing immediate feedback and monitoring, thereby improving the accuracy of the steering wheel control of the autonomous driving system.

[0036] Furthermore, based on the same concept, the present application also provides a vehicle-mounted terminal, which includes a processor and a memory. At least one instruction, at least one program, a code set or an instruction set is stored in the memory, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the above-mentioned steering wheel rotation angle display method.

[0037] Compared with the prior art, the beneficial effects of the present application are as follows:

[0038] The present application first initializes the angle calculation model and the corresponding reward function, and obtains the predicted steering wheel angle based on the angle calculation model and historical image data, so as to update the reward function based on the predicted steering wheel angle and the pre-stored actual steering wheel angle; when the values of the reward function all belong to a preset range after a preset number of iterations, it is determined that the angle calculation model training is completed. The angle calculation model trained by the present application can maintain high accuracy in different environments and conditions. Even in complex road conditions, the model can effectively respond and avoid the accuracy decline caused by environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Flow chart of the steering wheel angle calculation model training method described in this application.

[0040] Figure 2 Flow chart of the steering wheel rotation angle display method described in this application.

[0041] Figure 3 System framework diagram of the steering wheel rotation angle display described in this application. Detailed implementation manners

[0042] The following further describes in detail a steering wheel angle calculation model training method, a display method, a system and a terminal of this application in conjunction with specific embodiments and the accompanying drawings.

[0043] Please refer to Figure 1 , this application provides a steering wheel angle calculation model training method, including the following steps S1 - S2.

[0044] In some embodiments, it is first necessary to train an angle calculation model. Specifically: first design a reward function and an angle calculation model, then use the initial angle calculation model to extract image features from historical image data, and then obtain the predicted steering wheel angle based on the image features.

[0045] Furthermore, update the reward function and the angle calculation model based on the error between the obtained predicted steering wheel angle and the actual steering wheel angle, and then train the angle calculation model based on the updated reward function until the model converges, and deploy the trained angle calculation model to the in - vehicle terminal.

[0046] The following details steps S1 - S2.

[0047] Step S1: Initialize the angle calculation model and the corresponding reward function, and obtain the predicted steering wheel angle based on the angle calculation model and historical image data, so as to update the reward function based on the predicted steering wheel angle and the pre - stored actual steering wheel angle.

[0048] Furthermore, before initializing the angle calculation model, it includes:

[0049] Collect a number of historical image data of the steering wheel at different rotation angles in advance, and store the rotation angle corresponding to each historical image data as the actual steering wheel angle.

[0050] In some embodiments, collect historical image data containing different rotation angles through the built - in camera of the vehicle, annotate the rotation angle corresponding to each historical image data, and at the same time store the rotation angle of the annotated historical image data as the actual angle.

[0051] Among them, the historical image data is \(I\in\mathbb{R}\) H×W×C , where \(H\) is the image height, \(W\) is the image width, and \(C\) is the number of channels.

[0052] In the above technical solution, by collecting historical image data of the steering wheel at different rotation angles, the diversity of training data is ensured. This diversity helps the model to make accurate predictions at different rotation angles and avoids the overfitting problem that may be caused by a single-angle data set.

[0053] Furthermore, the initialization of the reward function includes:

[0054] Initialize the predicted angle of the steering wheel and the actual angle of the steering wheel, and initialize the reward function based on the difference between the initialized predicted angle of the steering wheel and the actual angle of the steering wheel.

[0055] In some embodiments, the reward function where \(\theta\) represents the predicted angle of the steering wheel, represents the actual angle of the steering wheel.

[0056] For example: first randomly initialize the predicted angle of the steering wheel to \(10^{\circ}\) and the actual angle of the steering wheel to \(15^{\circ}\). At this time, the initialized reward function \(r = - 5^{\circ}\). During model training, if the model can optimize the predicted angle to be close to the actual angle, the smaller the error, the greater the reward.

[0057] In the above technical solution, by calculating the difference between the initialized predicted angle of the steering wheel and the actual angle of the steering wheel, the error situation of the model in the initial state can be directly reflected. This initialization method based on error enables the reward function to be optimized for the actual performance of the model from the very beginning, avoiding the problem that the reward function is out of touch with the model performance.

[0058] Furthermore, the initialization of the angle calculation model includes:

[0059] Initialize the policy function, and initialize the angle calculation model based on the initialized policy function and the reward function.

[0060] In some embodiments, the angle calculation model where \(\pi\) is the policy function and \(R\) is the cumulative reward based on the reward function.

[0061] Among them, the initialization of the policy function includes: defining the state space \(s = I\), indicating that the input of the angle calculation model is image data; defining the action space, where the action \(a\in\mathbb{R}\) represents the predicted angle of the steering wheel, indicating that the task of the angle calculation model is to predict the rotation angle of the steering wheel according to the input image data.

[0062] In the above technical solution, by combining the policy function and the reward function, the angle calculation model adopts the framework of reinforcement learning. This framework allows the model to learn the optimal policy through interaction with the environment, rather than just learning based on static data, so as to better adapt to the dynamic environment.

[0063] Different from traditional supervised learning methods, reinforcement learning focuses on long-term cumulative rewards. Therefore, the model will consider the long-term impact of actions during the training process, which is particularly important for scenarios such as autonomous driving that require considering multiple future steps of operations.

[0064] Further, it also includes: initializing the accumulated reward based on the initialized reward function.

[0065] In the above technical solution, by initializing the accumulated reward based on the initialized reward function, it can directly reflect the performance level of the model in the initial state, which helps to evaluate the initial performance of the model and provide a benchmark for subsequent training and optimization.

[0066] Further, the angle calculation model at least adopts a convolutional neural network model, and the convolutional neural network model at least includes a convolutional layer and a fully connected layer.

[0067] The obtaining of the predicted steering wheel angle includes:

[0068] Inputting the historical image data into the initialized or updated angle calculation model to extract the image features corresponding to the historical image data based on the convolutional layer, and mapping the image features to the predicted steering wheel angle based on the fully connected layer.

[0069] In some embodiments, the angle calculation model extracts image features based on a convolutional neural network, including: performing a convolution operation on the input historical image data through a convolutional kernel to extract local features in the image, such as edge features, texture features, etc.; and further mapping the image features extracted by the convolutional kernel to the final prediction result, that is, the predicted steering wheel angle.

[0070] In the above technical solution, the unique structure of the convolutional layer can efficiently extract local features (such as edges, textures, shapes, etc.) and global features in the image. Through multiple convolutional operations, the model can gradually extract high-level features (contours, structures) from low-level features (edges, lines), thereby more comprehensively understanding the features of the steering wheel image. This efficient feature extraction ability enables the model to more accurately identify the rotation angle of the steering wheel, especially performing well in complex image backgrounds; the fully connected layer can map the rich features extracted by the convolutional layer to the specific predicted angle of the steering wheel through a non-linear activation function (such as ReLU) and weight adjustment. The powerful mapping ability of the fully connected layer enables the model to handle complex angle prediction tasks and continuously approach the real rotation angle of the steering wheel by optimizing the weight parameters.

[0071] Further, after obtaining the predicted angle of the steering wheel output by the angle calculation model, obtain the corresponding actual angle of the steering wheel pre-stored in the historical image data, so as to calculate a new reward function based on the predicted angle of the steering wheel and the actual angle of the steering wheel, and then update the reward function.

[0072] Further, after updating the reward function, it includes:

[0073] Update the cumulative reward based on the updated reward function.

[0074] Calculate the policy gradient based on the updated reward function, the current policy function, and the cumulative reward, and update the policy function according to the policy gradient.

[0075] And update the angle calculation model based on the updated reward function and policy function.

[0076] In some embodiments, in reinforcement learning, the reward is accumulated step by step. In order to consider the discount effect of future rewards, a discount factor (i.e., the reward function) is used to calculate the cumulative reward starting from the current time.

[0077] Further, the core of the policy gradient method is to optimize the policy function through the gradient ascent method to maximize the cumulative reward. The policy gradient is And then update the policy function parameter θ through the gradient ascent method.

[0078] where α is the learning rate, which controls the step size of the update of the policy function parameters. When the learning rate is large, the parameter update is fast, but it may lead to unstable training; when the learning rate is small, the parameter update is slow, but the training process is more stable.

[0079] Further, due to the policy function π θ(a|s) and the angle calculation model are the same neural network. By updating the parameters θ of the policy function, the parameters of the angle calculation model are indirectly updated. The updated angle calculation model can generate more accurate prediction results.

[0080] Furthermore, the updated angle calculation model will be used in the next training iteration to continue extracting image features from the input historical image data and generating new predicted steering wheel angles. By continuously optimizing the reward function and the policy function, the prediction performance of the model will be gradually improved.

[0081] In the above technical solution, by calculating the policy gradient based on the updated reward function, the current policy function, and the accumulated reward, and updating the policy function according to the policy gradient, the model can continuously optimize its policy during the training process. This gradient-based update method can make the model gradually tend to the optimal policy. Dynamically updating the reward function and the policy function can reduce the risk of overfitting of the model. Overfitting usually occurs when the model over-adapts to the training data and loses its generalization ability. By continuous optimization and update, the model can better adapt to new data and new environments.

[0082] Step S2: When the values of the reward function all belong to a preset range after a preset number of iterations, it is determined that the training of the angle calculation model is completed.

[0083] In some embodiments, a preset number of iterations is set, which represents the number of iterations in a training cycle, and a preset range of reward function values (i.e., the preset range) is set to represent the expected range of the values of the reward function.

[0084] A list can also be initialized to store the reward function values of the most recent N iterations.

[0085] After each training cycle, check whether the values of the reward function in the list are all within the preset range. If the above conditions are met, it is determined that the training of the angle calculation model is completed and the training loop is exited. If the above conditions are not met, the iteration of the next training cycle needs to be continued.

[0086] In other embodiments, an error reporting program can also be set during the model training process. If the model still does not converge after multiple training cycles, an error message is output to the administrator or technical developer, etc., to promptly discover and correct the problem.

[0087] Furthermore, based on the same concept, please refer to Figure 2, this application also provides a method for displaying the steering wheel rotation angle, including: obtaining the steering wheel image in real time, and inputting the steering wheel image into the angle calculation model to output and display the steering wheel rotation angle information; wherein, the angle calculation model is obtained by training through the above-mentioned steering wheel angle calculation model training method.

[0088] In some embodiments, the trained angle calculation model is deployed on the in-vehicle terminal, and the in-vehicle terminal is connected to the vehicle's built-in camera and display system through a serial port.

[0089] After the built-in camera obtains the steering wheel image, it is input into the angle calculation model of the in-vehicle terminal through the serial port, so as to output the calculated steering wheel rotation angle information in real time based on the obtained steering wheel image through the angle calculation model, and the steering wheel rotation angle information is output to the vehicle's built-in display screen or HUD through the serial port.

[0090] In other embodiments, after the steering wheel image is collected, the steering wheel image can be preprocessed to make the steering wheel image meet the input requirements of the model, and the preprocessing includes normalization, size adjustment, channel conversion, etc.

[0091] It should also be noted that the in-vehicle terminal and the camera and display system are not necessarily connected through a serial port, and those skilled in the art can select other connection methods according to actual application requirements, which are not limited thereto.

[0092] Furthermore, based on the same concept, please refer to Figure 3 , this application also provides a steering wheel rotation angle display system, including:

[0093] The acquisition module is used to obtain the steering wheel image in real time and send the steering wheel image to the calculation module.

[0094] The calculation module is deployed with the angle calculation model to calculate the steering wheel rotation angle information based on the angle calculation model according to the steering wheel image, and output the steering wheel rotation angle information to the display module.

[0095] And the display module is used to display the steering wheel rotation angle information.

[0096] In some embodiments, the acquisition module is the vehicle's built-in camera, which is used to collect the steering wheel image in real time and input it into the calculation module (equivalent to the in-vehicle terminal) through the serial port; the calculation module calculates the steering wheel rotation angle information based on the collected steering wheel image and outputs it to the display module (such as the vehicle's built-in display screen or HUD) for display through the serial port.

[0097] It should be noted that the specific implementation of the training method of the angle calculation model has been described in various embodiments of the steering wheel angle calculation model training method, and will not be elaborated here.

[0098] In the above technical solution, the acquisition module can obtain the steering wheel image in real time to ensure that the system can timely capture the dynamic changes of the steering wheel; the calculation module deploys an efficient angle calculation model (such as a convolutional neural network), which can complete image feature extraction and angle calculation in a short time to ensure the real-time output of the rotation angle information; the display module can display the calculated steering wheel rotation angle information in real time, providing instant feedback and monitoring, thereby improving the accuracy of the steering wheel control of the autonomous driving system.

[0099] Furthermore, based on the same concept, the present application also provides a vehicle-mounted terminal, which includes a processor and a memory. At least one instruction, at least one program, a code set or an instruction set is stored in the memory, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the steering wheel angle calculation model training method described above.

[0100] In some embodiments, the memory and the processor are interconnected through a bus; the processor can be one or more CPUs. When the processor is a single CPU, the CPU can be a single-core CPU or a multi-core CPU, and the processor is used to control each functional module of the vehicle-mounted terminal and process signals.

[0101] The memory includes but is not limited to RAM (Random Access Memory), ROM (Read-Only Memory), EPROM (Erasable Programmable Read-Only Memory), CD-ROM (Compact Disc Read-Only Memory). The memory is used to store computer programs, operating systems, various applications and data, such as storing the computer program for implementing the steering wheel angle calculation model training method.

[0102] In summary, the present application provides a method for training a steering wheel angle calculation model, a display method, a system, and a terminal. First, the angle calculation model and the corresponding reward function are initialized, and the predicted steering wheel angle is obtained based on the angle calculation model and historical image data, so as to update the reward function based on the predicted steering wheel angle and the actually stored steering wheel angle. When the values corresponding to the reward function all fall within a preset range after a preset number of iterations, it is determined that the training of the angle calculation model is completed. The angle calculation model trained by the present application can maintain high accuracy under different environments and conditions. Even in complex road conditions, the model can effectively cope with them, avoiding the accuracy decline caused by environmental changes.

[0103] Although example embodiments have been described herein with reference to the drawings, it should be understood that the above example embodiments are merely exemplary and are not intended to limit the scope of the present application thereto. Those of ordinary skill in the art can make various changes and modifications therein without departing from the scope and spirit of the present application. All such changes and modifications are intended to be included within the scope of the present application as claimed in the appended claims.

[0104] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0105] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0106] Each component embodiment of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) can be used in practice to implement some or all of the functions of some modules according to the embodiments of the present application. The present application can also be implemented as a device program (for example, a computer program and a computer program product) for executing part or all of the methods described herein. Such a program for implementing the present application can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0107] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0108] Although the description of the present application is made in conjunction with the above specific embodiments, it is obvious that those skilled in the art can make many substitutions, modifications and variations based on the above content. Therefore, all such substitutions, improvements and variations are included within the spirit and scope of the appended claims.

Claims

1. A steering wheel angle calculation model training method, characterized in that: include: Initializing an angle calculation model and a corresponding reward function, and obtaining a predicted steering wheel angle based on the angle calculation model and historical image data, so as to update the reward function based on the predicted steering wheel angle and a pre-stored actual steering wheel angle; When the reward function has been iterated a preset number of times and the values ​​corresponding to the reward function are all within a preset range, it is determined that the angle calculation model training is completed.

2. The steering wheel angle calculation model training method according to claim 1, characterized in that: Before initializing the angle calculation model, include: A plurality of historical image data of the steering wheel at different rotation angles are collected in advance, and the rotation angle corresponding to each historical image data is stored as the actual angle of the steering wheel.

3. The steering wheel angle calculation model training method according to claim 2, characterized in that: The initialization reward function includes: Initialize the predicted steering wheel angle and the actual steering wheel angle to initialize the reward function based on the difference between the initialized predicted steering wheel angle and the actual steering wheel angle.

4. The steering wheel angle calculation model training method according to claim 3, characterized in that: The initialization angle calculation model includes: Initialize the policy function to initialize the angle calculation model based on the initialized policy function and reward function.

5. The steering wheel angle calculation model training method according to claim 4, characterized in that: The angle calculation model at least adopts a convolutional neural network model, and the convolutional neural network model at least includes a convolutional layer and a fully connected layer; The obtaining of the predicted steering wheel angle comprises: The historical image data is input into an initialized or updated angle calculation model to extract image features corresponding to the historical image data based on the convolution layer, and the image features are mapped to a predicted steering wheel angle based on the fully connected layer.

6. The steering wheel angle calculation model training method according to claim 4, characterized in that: Also includes: Initialize the accumulated reward based on the initialized reward function.

7. The steering wheel angle calculation model training method according to claim 6, characterized in that: After updating the reward function, it includes: Updating the accumulated reward based on the updated reward function; Calculating a policy gradient based on the updated reward function and the current policy function and the accumulated reward, and updating the policy function according to the policy gradient; And, updating the angle calculation model based on the updated reward function and policy function.

8. A method for displaying a steering wheel rotation angle, characterized in that: include: Acquire a steering wheel image in real time, and input the steering wheel image into an angle calculation model to output and display steering wheel rotation angle information; Wherein, the angle calculation model is obtained by training according to the steering wheel angle calculation model training method as described in any one of claims 1-7.

9. A steering wheel rotation angle display system, characterized in that: include: An acquisition module, used for acquiring a steering wheel image in real time and sending the steering wheel image to a calculation module; The calculation module is deployed with the angle calculation model as claimed in any one of claims 1 to 7, so as to calculate the steering wheel rotation angle information according to the steering wheel image based on the angle calculation model, and output the steering wheel rotation angle information to the display module; And the display module is used to display the steering wheel rotation angle information.

10. A vehicle-mounted terminal, characterized in that: The vehicle-mounted terminal includes a processor and a memory, and the memory stores at least one instruction, at least one program, a code set or an instruction set. The at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the steering wheel rotation angle display method as described in claim 8.

Images