Method and system for evaluating and correcting automobile driving behavior

Abstract

The invention discloses a method and system for evaluating and correcting automobile driving behaviors, and the method comprises the steps: integrating the data of an internal sensor and an external sensor of an automobile, and designing a self-adaptive machine learning algorithm to recognize an individual driving style. Then, a complex driving behavior mode is identified by using a model of a hybrid convolutional neural network and a recurrent neural network. And finally, generating a personalized driving behavior correction scheme through a generative adversarial network technology. In addition, the invention further comprises a real-time augmented reality feedback and interaction system, and the system can provide necessary driving information and suggestions in real time through a head-up display or AR glasses, which is helpful for the driver to improve the driving safety and convenience, and greatly improves the driving experience at the same time.

Description

Technical Field

[0001] The present invention belongs to the field of driver behavior assessment, and more specifically relates to a method and system for assessing and correcting automobile driving behavior. Background Art

[0002] With the rapid development of science and technology, cars are gradually equipped with more intelligent features. However, the problem that comes with it is that the matching degree between the driver's driving behavior and these intelligent features is often not ideal. This includes but is not limited to the driver's overly aggressive or conservative operation and distracted vision, which may lead to reduced driving safety and poor driving experience. In addition, when facing a complex driving environment, whether the driver's judgment and operation comply with driving rules and whether there is a certain degree of safety guarantee is also part of the vehicle evaluation.

[0003] Traditional driving behavior assessment methods are usually based on the driver's driving trajectory, speed, acceleration, braking, etc., and the driver's driving behavior and driving skills are evaluated through professional observation and analysis. However, this method cannot implement microscopic and real-time driving behavior assessment and correction, and it is also unable to perform personalized assessment and correction based on the individual characteristics of the driver.

[0004] In addition, although the existing technical solutions have added assisted driving functions to the vehicle system, such as lane keeping, automatic parking, automatic cruising, etc., these systems are more inclined to correct and track the driver's operations during driving, rather than evaluate and correct the driver's driving habits, and cannot fundamentally change certain bad driving habits of the driver.

[0005] Therefore, it is necessary to develop a system that can evaluate and correct the driver's driving behavior in real time to improve driving safety and the driver's driving skills. At the same time, with the help of modern computer vision, machine learning and other technologies, combined with augmented reality (AR) technology, a more intuitive and natural interface and interaction method can be provided to further improve the user experience. Summary of the invention

[0006] The technical problem to be solved by the present invention is how to effectively identify and evaluate the driver's driving behavior, and how to generate a personalized driving behavior correction plan. At the same time, considering the requirements of real-time, interactivity and reliability, the present invention also needs to solve how to build a real-time feedback and interaction system based on augmented reality (AR), so that the driver can obtain necessary driving information and suggestions while maintaining focus on driving, thereby improving driving safety and convenience.

[0007] In order to achieve the above object, the present invention is implemented by adopting the following technical scheme: the method comprises:

[0008] Data collection: Integrate the vehicle's internal sensor data with external environment sensor data, and use edge computing technology to perform preliminary data processing locally to reduce data transmission volume and latency.

[0009] Design an adaptive machine learning algorithm to automatically adjust the parameters of the analysis model according to the driving habits of different drivers; identify individual driving styles;

[0010] Driving behavior pattern recognition: Identify complex driving behavior patterns by building a hybrid model of convolutional neural networks and recurrent neural networks;

[0011] Personalized behavior correction plan generation: Generate personalized driving behavior correction plans using generative adversarial network technology.

[0012] In one embodiment, the data acquisition uses Kalman filtering to fuse multi-sensor data;

[0013] The edge computing uses a principal component analysis algorithm to achieve data compression and feature extraction to reduce data dimension and transmission volume.

[0014] In one embodiment, the adaptive machine learning algorithm includes:

[0015] Initial model training, use the pre-labeled driving behavior dataset to train the initial model and select the support vector machine algorithm;

[0016] Driving style recognition, using the clustering algorithm K-means to divide driving data into different categories; assuming that each cluster represents a specific driving style;

[0017]

[0018] in, Si is the point set of cluster i, μ i is the centroid of cluster i;

[0019] Adaptive model adjustment uses the online gradient descent method to dynamically adjust model parameters; assuming that the loss function of the current model is L(θ), where θ is the model parameter; at each time step, the parameters are updated by minimizing the loss function:

[0020]

[0021] Where η is the learning rate, is the gradient of the loss function.

[0022] In one embodiment, the driving behavior pattern recognition is performed by constructing a hybrid model of a convolutional neural network (CNN) and a recurrent neural network (RNN);

[0023] First, convolutional neural networks are used to extract spatiotemporal features in driving behavior pattern recognition; local features are extracted from the input data through the convolution layer, which is expressed as:

[0024]

[0025] in, is the output of the kth convolution kernel at position (i, j), x i+m,j+n is a local region of the input data, is the weight of the convolution kernel, b k is the bias and f is the activation function.

[0026] Next, the recurrent neural network (RNN) processes time series data and is therefore used to capture time-dependent features in driving behavior pattern recognition. RNN memorizes sequence information through its hidden state, and the RNN unit is represented as:

[0027] h t =σ(W h x t +U h h t-1 +b h )

[0028] o t =V h h t +c

[0029] Among them, h t is the hidden state at time step t, x t is the input, W h , U h is the weight matrix, b h is the bias, σ is the activation function, o t is the output, V h and c are the weight and bias of the output layer respectively; RNN can capture the temporal dynamic characteristics of driving behavior;

[0030] Combining CNN and RNN, a hybrid model is constructed. First, CNN extracts features from driving behavior data, and then RNN performs time series modeling on the extracted features. During the training process, the back propagation algorithm is used to optimize the model parameters by minimizing the loss function. The loss function is expressed as:

[0031]

[0032] Where L is the loss, N is the number of samples, and y i is the true label, is the predicted probability.

[0033] In one embodiment, the personalized driving behavior correction program includes:

[0034] First, the generator G receives a random noise vector z as input and generates a sample G(z) that simulates driving behavior through a series of nonlinear transformations. The goal of the generator is to generate data close to real driving behavior to deceive the discriminator. The output of the generator is expressed as:

[0035] G(z)=f G (z;θ G )

[0036] Among them, f G is the mapping function of the generator, θ G are the parameters of the generator;

[0037] The discriminator D receives a sample as input and outputs a scalar, which indicates the probability that the sample is real data. The goal of the discriminator is to correctly distinguish between real driving behavior samples (from training data) and generated samples G(z). The output of the discriminator is expressed as:

[0038] D(x)=f D (x;θ D )

[0039] Among them, f D is the mapping function of the discriminator, θ D are the parameters of the discriminator.

[0040] The training process of GAN is a minimax game. The generator and discriminator optimize their parameters through adversarial training. The objective function is expressed as:

[0041] min G max D V(D,G)=E x～pdata(x) [logD(x)]+E z～pz(z) [log(1-D(G(z)))]

[0042] Among them, p data (x) is the real data distribution, p z (z) is the noise vector distribution, V(D,G) is the objective function, E x～pdata(x) The expected value of the true data distribution, E z～pz(z) is the expected value of the noise vector distribution.

[0043] In another aspect, a system for evaluating and correcting automobile driving behavior is provided, wherein the system is applicable to the method described above, and the system comprises:

[0044] Data acquisition module: collects data from vehicle internal sensors and external environment sensors, uses Kalman filtering to perform multi-sensor data fusion, and uses principal component analysis to perform data compression and feature extraction;

[0045] Adaptive machine learning algorithm module: Use the support vector machine algorithm to train the pre-labeled driving behavior data set, use the K-means clustering algorithm to classify the driving data into different driving style categories, and use the online gradient descent method to dynamically adjust the model parameters;

[0046] Driving behavior pattern recognition module: extracts the spatiotemporal features of driving behavior data through convolutional neural networks, uses recurrent neural networks to perform time series modeling on the extracted features, and uses back propagation algorithms to optimize model parameters;

[0047] Personalized behavior correction solution generation module: Combined with generative adversarial network technology, the random noise vector input generator generates samples close to real driving behavior, while the discriminator distinguishes between real driving behavior samples and generated samples, and optimizes their respective parameters through adversarial training;

[0048] Augmented reality (AR) feedback and interaction system: This system is installed on the head-up display HUD or AR glasses, processes the vehicle's sensor data in real time, and combines preset map data and traffic information to generate suggestions and warning information related to the current driving situation.

[0049] Beneficial effects of the present invention:

[0050] By analyzing the driver's driving behavior data and environmental data, identifying patterns and generating personalized driving behavior correction plans, it helps improve driving safety and reduce the occurrence of traffic accidents. Real-time driving behavior evaluation and feedback can be achieved, which can immediately correct the driver's driving habits and encourage the driver to form safer and more traffic-compliant driving habits. Using augmented reality (AR) to intuitively display driving advice, warnings and other information in the driver's field of view, it will not distract the driver's attention and ensure driving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flow chart of the method of the present invention;

[0052] Figure 2 A hybrid model flow chart of a convolutional neural network (CNN) and a recurrent neural network (RNN) is constructed for the present invention.

[0053] Figure 3 Block diagram of the system of the present invention. DETAILED DESCRIPTION

[0054] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0055] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used in the present invention in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0056] Embodiment 1:

[0057] like Figure 1 As shown, the method and system for evaluating and correcting automobile driving behavior include:

[0058] S1. Data collection: Integrate the vehicle’s internal sensor data with external environment sensor data, and use edge computing technology to perform preliminary data processing locally to reduce data transmission volume and latency.

[0059] Integrate the vehicle's internal sensors (such as OBD-II interface) and external environmental sensors (such as LiDAR, camera) into a unified hardware platform. Design a central control unit (MCU or FPGA) responsible for the acquisition and preliminary processing of sensor data. Ensure that the clocks of each sensor are synchronized to ensure the time consistency of the data. This can be achieved through the Network Time Protocol (NTP) or a dedicated synchronization signal.

[0060] Kalman filter is used to fuse multi-sensor data.

[0061] Deploy edge computing capabilities on local hardware, using microprocessors or embedded GPUs for preliminary processing and analysis of data. This can significantly reduce the need to transmit data to the cloud and reduce latency.

[0062] Implement data compression and feature extraction algorithms, such as principal component analysis (PCA) or discrete Fourier transform (DFT), to reduce data dimensionality and transmission volume.

[0063] Through the above steps, the intelligent data acquisition module can effectively fuse data from different sensors and perform preliminary processing locally. This not only improves the efficiency of data processing, but also provides high-quality input data for subsequent driving behavior analysis.

[0064] S2. Design an adaptive machine learning algorithm to automatically adjust the parameters of the analysis model according to the driving habits of different drivers; identify individual driving styles.

[0065] First, data related to driving behavior is collected. This may include sensor data such as speed, acceleration, braking, steering angle, etc. The data needs to go through pre-processing steps such as denoising, normalization, and feature extraction.

[0066] Extract features from raw data. You can use the time window method to extract statistical features, such as mean, variance, maximum, minimum, etc.

[0067] Initial model training, use the pre-labeled driving behavior dataset to train the initial model and select the support vector machine algorithm;

[0068] Driving style recognition, using the clustering algorithm K-means to divide driving data into different categories; assuming that each cluster represents a specific driving style;

[0069]

[0070] Among them, S i is the point set of cluster i, μ i is the centroid of cluster i;

[0071] Adaptive model adjustment uses the online gradient descent method to dynamically adjust model parameters; assuming that the loss function of the current model is L(θ), where θ is the model parameter; at each time step, the parameters are updated by minimizing the loss function:

[0072]

[0073] Where η is the learning rate, is the gradient of the loss function.

[0074] S3. Driving behavior pattern recognition: Identify complex driving behavior patterns by building a hybrid model of convolutional neural network and recurrent neural network.

[0075] In driving behavior pattern recognition, the introduction of deep learning technology provides new possibilities for processing and analyzing complex driving behavior data. By constructing a hybrid model of convolutional neural network (CNN) and recurrent neural network (RNN), the driver's behavior characteristics in different environments and situations can be effectively captured.

[0076] The driving behavior pattern recognition is achieved by constructing a hybrid model of a convolutional neural network (CNN) and a recurrent neural network (RNN);

[0077] S301. First, the convolutional neural network is used to extract spatiotemporal features in driving behavior pattern recognition. Local features of the input data are extracted through the convolution layer, which is expressed as:

[0078]

[0079] in, is the output of the kth convolution kernel at position (i, j), x i+m,j+n is a local region of the input data, is the weight of the convolution kernel, b k is the bias and f is the activation function. In this way, CNN can identify local patterns in driving behavior data, such as acceleration changes and direction changes.

[0080] S302, Next, the recurrent neural network (RNN) processes time series data and is therefore used to capture time-dependent features in driving behavior pattern recognition. RNN memorizes sequence information through its hidden state, and the RNN unit is represented as:

[0081] h t =σ(W h x t +U h h t-1 +b h )

[0082] o t =V h h t +c

[0083] Among them, h t is the hidden state at time step t, x t is the input, W h , U h is the weight matrix, b h is the bias, σ is the activation function, o t is the output, V h and c are the weight and bias of the output layer respectively; RNN can capture the temporal dynamic characteristics of driving behavior, such as the driver's acceleration and braking patterns.

[0084] S303. By combining CNN and RNN, a hybrid model can be constructed. First, CNN extracts features from driving behavior data, and then RNN performs time series modeling on the extracted features. This combination can process spatial and temporal features at the same time, providing a more comprehensive understanding of driving behavior. During the training process, the back propagation algorithm is used to optimize the model parameters by minimizing the loss function. The loss function is usually the cross entropy loss (for classification tasks) or the mean square error (for regression tasks), expressed as:

[0085]

[0086] Where L is the loss, N is the number of samples, and y i is the true label, is the predicted probability.

[0087] Through this deep learning method, the driving behavior pattern recognition system can more accurately understand and predict the driver's behavior, improving driving safety and comfort.

[0088] S4. Generation of personalized behavior correction plan: Generate personalized driving behavior correction plan using generative adversarial network technology.

[0089] In the process of generating personalized driving behavior correction solutions, Generative Adversarial Network (GAN) technology provides an innovative method to help drivers improve their driving habits by generating realistic simulated environments and behavioral suggestions. GAN consists of two main components: Generator and Discriminator, which are trained in an adversarial manner.

[0090] First, the generator G receives a random noise vector z as input and generates a sample G(z) that simulates driving behavior through a series of nonlinear transformations. The goal of the generator is to generate data close to real driving behavior to deceive the discriminator. The output of the generator is expressed as:

[0091] G(z)=f G (z;θ G )

[0092] Among them, f G is the mapping function of the generator, θ G are the parameters of the generator;

[0093] The discriminator D receives a sample as input and outputs a scalar, which indicates the probability that the sample is real data. The goal of the discriminator is to correctly distinguish between real driving behavior samples (from training data) and generated samples G(z). The output of the discriminator is expressed as:

[0094] D(x)=fD (x;θ D )

[0095] Among them, f D is the mapping function of the discriminator, θ D are the parameters of the discriminator.

[0096] The training process of GAN is a minimax game. The generator and discriminator optimize their parameters through adversarial training. The objective function is expressed as:

[0097]

[0098] Among them, p data (x) is the real data distribution, p z (z) is the noise vector distribution, V(D,G) is the objective function, The expected value of the true data distribution, is the expected value of the noise vector distribution.

[0099] Through this process, GAN is able to generate realistic driving behavior samples for creating personalized driving behavior correction plans. These plans not only contain specific improvement suggestions, but can also provide an immersive simulation training environment by combining with virtual reality (VR) technology. In this environment, drivers can experience and practice improved driving strategies, thereby improving their actual driving skills.

[0100] To achieve this, the generated driving behavior samples are used as input to create a dynamic VR simulation scene. The driver enters this scene through a VR device, and the system adjusts the simulation environment and feedback in real time based on their current driving behavior and the generated correction plan. This interactive training method not only helps the driver understand the improvement suggestions, but also enhances their mastery of new strategies through practice.

[0101] Through the combination of GAN and VR technology, personalized behavior correction solutions can provide an innovative and effective driving training method, helping drivers to improve themselves in a safe virtual environment and ultimately improve their performance in real driving scenarios.

[0102] Embodiment 2

[0103] like Figure 3 As shown, a system for evaluating and correcting automobile driving behavior, the system is applicable to the method described, and the system comprises:

[0104] Data acquisition module: collects data from vehicle internal sensors and external environment sensors, uses Kalman filtering to perform multi-sensor data fusion, and uses principal component analysis to perform data compression and feature extraction.

[0105] Adaptive machine learning algorithm module: Use the support vector machine algorithm to train the pre-labeled driving behavior dataset, use the K-means clustering algorithm to classify the driving data into categories with different driving styles, and use the online gradient descent method to dynamically adjust the model parameters.

[0106] Driving behavior pattern recognition module: The spatiotemporal features of driving behavior data are extracted through convolutional neural networks, and recurrent neural networks are used to perform time series modeling on the extracted features, and the back propagation algorithm is used to optimize the model parameters.

[0107] Personalized behavior correction plan generation module: Combined with generative adversarial network technology, the random noise vector input generator is used to generate samples close to real driving behavior, while the discriminator distinguishes between real driving behavior samples and generated samples, and optimizes their respective parameters through adversarial training.

[0108] Augmented reality (AR) feedback and interaction system: This system is installed on the head-up display HUD or AR glasses, processes the vehicle's sensor data in real time, and combines preset map data and traffic information to generate suggestions and warning information related to the current driving situation.

[0109] When designing a real-time feedback and interaction system based on augmented reality (AR), the goal is to improve driving safety and user experience through innovative technology. The core of this system is to project driving advice and warning information directly into the driver's field of vision, using AR technology to enhance the intuitiveness and timeliness of information acquisition.

[0110] First, the AR system requires a high-precision display device, such as a head-up display (HUD) mounted on the windshield or AR glasses. These devices can overlay digital information in the driver's natural field of view without blocking the view of the road. To achieve this, the system must process the vehicle's sensor data, such as speed, direction, GPS location, etc. in real time, and combine it with preset map data and traffic information to generate suggestions and warnings related to the current driving situation. After processing, this information is projected into the driver's field of view through the AR device, allowing him to obtain important information without diverting his attention.

[0111] Next, the system needs to integrate voice assistants and gesture recognition technology to achieve natural human-computer interaction. The voice assistant uses natural language processing (NLP) technology to recognize and understand the driver's voice commands and provide corresponding feedback or perform corresponding operations. For example, the driver can request navigation information or adjust system settings by voice. Gesture recognition technology uses in-car cameras and computer vision algorithms to recognize the driver's gestures to achieve certain shortcut operations, such as answering calls or adjusting the volume. This multimodal interaction method is designed to minimize interference with the driver's attention.

[0112] During the system development process, ensuring low latency and high reliability is key. Real-time requires the system to respond quickly to driver input and environmental changes, so data processing and transmission speeds need to be optimized. The system's software architecture should adopt efficient algorithms and data structures, and use edge computing technology to distribute some computing tasks to on-board devices to reduce dependence on the cloud. In addition, the system's user interface design must be concise and clear to avoid information overload, while ensuring visibility under various lighting conditions.

[0113] Through this AR-based real-time feedback and interactive system, drivers can obtain necessary driving information and suggestions while staying focused on driving, thereby improving driving safety and convenience. This innovative design not only improves the driving experience, but also lays the foundation for the development of future intelligent driving technology.

[0114] This application proposes a method of using a hybrid model, including a convolutional neural network (CNN) and a recurrent neural network (RNN) to identify complex driving behavior patterns. By constructing a hybrid model of CNN and RNN, the spatiotemporal characteristics and temporal dependencies of driving behavior can be effectively captured. Furthermore, this application proposes a method of using a generative adversarial network (GAN) to generate personalized driving behavior correction solutions. The generative adversarial network consists of a generator and a discriminator. The goal of the generator is to generate data that is close to real driving behavior to deceive the discriminator, while the goal of the discriminator is to effectively distinguish between real driving behavior samples and generated samples. Finally, in conjunction with augmented reality (AR) technology, the virtual head-up display HUD and AR glasses can provide real-time feedback of the correction solution to the driver.

[0115] The present invention realizes effective recognition of driving behavior by designing a hybrid model of convolutional neural network (CNN) and recurrent neural network (RNN). Among them, CNN is used to extract spatiotemporal features, and RNN is used to capture time-dependent features, which can more effectively capture patterns and trends in driving behavior. A generative adversarial network (GAN) is used to generate personalized driving behavior correction schemes. The generative adversarial network consists of a generator and a discriminator. The driving behavior data generated by the generator can be as close to the real driving behavior data as possible, and the goal of the discriminator is to distinguish the generated driving behavior data from the real driving behavior data. Through this adversarial method, a driving behavior that can realistically simulate the driver's driving style can be generated, providing the driver with a correction scheme that is closer to his driving habits.

[0116] It also combines augmented reality (AR) technology, which can provide real-time feedback of correction plans to the driver through a virtual head-up display (HUD) or AR glasses. This feedback method is more intuitive, helping the driver to understand and accept the correction plan, thereby better correcting his driving behavior.

[0117] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0118] It should be understood that the detailed description of the technical solutions of the present invention by means of the preferred embodiments is illustrative rather than restrictive. A person skilled in the art may modify the technical solutions described in the embodiments, or replace some of the technical features by equivalents, based on reading the specification of the present invention; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating and correcting vehicle driving behavior, characterized in that: The method includes: Data collection: Integrate the vehicle’s internal sensor data with external environment sensor data, and use edge computing technology to perform preliminary data processing locally to reduce data transmission volume and latency; Design an adaptive machine learning algorithm to automatically adjust the parameters of the analysis model according to the driving habits of different drivers; identify individual driving styles; Driving behavior pattern recognition: Identify complex driving behavior patterns by building a hybrid model of convolutional neural networks and recurrent neural networks; Personalized behavior correction plan generation: Generate personalized driving behavior correction plans using generative adversarial network technology.

2. A method for evaluating and correcting vehicle driving behavior according to claim 1, characterized in that: The data acquisition uses Kalman filtering to fuse multi-sensor data; The edge computing uses a principal component analysis algorithm to achieve data compression and feature extraction to reduce data dimension and transmission volume.

3. The method for evaluating and correcting vehicle driving behavior according to claim 1, characterized in that: The adaptive machine learning algorithm includes: Initial model training, use the pre-labeled driving behavior dataset to train the initial model and select the support vector machine algorithm; Driving style recognition, using the clustering algorithm K-means to divide driving data into different categories; assuming that each cluster represents a specific driving style; in, Si is the point set of cluster i, μ i is the centroid of cluster i; Adaptive model adjustment uses the online gradient descent method to dynamically adjust model parameters; assuming that the loss function of the current model is L(θ), where θ is the model parameter; at each time step, the parameters are updated by minimizing the loss function: Where η is the learning rate, is the gradient of the loss function.

4. The method for evaluating and correcting vehicle driving behavior according to claim 1, characterized in that: The driving behavior pattern recognition is achieved by constructing a hybrid model of a convolutional neural network (CNN) and a recurrent neural network (RNN); First, convolutional neural networks are used to extract spatiotemporal features in driving behavior pattern recognition; local features are extracted from the input data through the convolution layer, which is expressed as: in, is the output of the kth convolution kernel at position (i, j), x i+m,j+n is a local region of the input data, is the weight of the convolution kernel, b k is the bias, f is the activation function; Next, the recurrent neural network (RNN) processes time series data and is therefore used to capture time-dependent features in driving behavior pattern recognition. The RNN memorizes sequence information through its hidden state. The RNN unit is represented as: h t =σ(W h x t +U h h t-1 +b h ) about t =V h h t +c Among them, h t is the hidden state at time step t, x t is the input, W h , U h is the weight matrix, b h is the bias, σ is the activation function, o t is the output, V h and c are the weight and bias of the output layer respectively; RNN can capture the temporal dynamic characteristics of driving behavior; Combining CNN and RNN, a hybrid model is constructed. First, CNN extracts features from driving behavior data, and then RNN performs time series modeling on the extracted features. During the training process, the back propagation algorithm is used to optimize the model parameters by minimizing the loss function. The loss function is expressed as: Where L is the loss, N is the number of samples, and y i is the true label, is the predicted probability.

5. The method for evaluating and correcting vehicle driving behavior according to claim 1, characterized in that: The personalized driving behavior correction program includes: First, the generator G receives a random noise vector z as input and generates a sample G(z) that simulates driving behavior through a series of nonlinear transformations. The goal of the generator is to generate data close to real driving behavior to deceive the discriminator. The output of the generator is expressed as: G(z)=f G (z;θ G ) Among them, f G is the mapping function of the generator, θ G are the parameters of the generator; The discriminator D receives a sample as input and outputs a scalar, which indicates the probability that the sample is real data. The goal of the discriminator is to correctly distinguish between real driving behavior samples (from training data) and generated samples G(z). The output of the discriminator is expressed as: D(x)=f D (x;θ D ) Among them, f D is the mapping function of the discriminator, θ D are the parameters of the discriminator; The training process of GAN is a minimax game. The generator and discriminator optimize their parameters through adversarial training. The objective function is expressed as: Among them, p data (x) is the real data distribution, p z (z) is the noise vector distribution, V(D,G) is the objective function, The expected value of the true data distribution, is the expected value of the noise vector distribution.

6. A system for evaluating and correcting automobile driving behavior, the system being applicable to the method according to any one of claims 1 to 5, characterized in that: The system comprises: Data acquisition module: collects data from vehicle internal sensors and external environment sensors, uses Kalman filtering to perform multi-sensor data fusion, and uses principal component analysis to perform data compression and feature extraction; Adaptive machine learning algorithm module: Use the support vector machine algorithm to train the pre-labeled driving behavior data set, use the K-means clustering algorithm to classify the driving data into different driving style categories, and use the online gradient descent method to dynamically adjust the model parameters; Driving behavior pattern recognition module: extracts the spatiotemporal features of driving behavior data through convolutional neural networks, uses recurrent neural networks to perform time series modeling on the extracted features, and uses back propagation algorithms to optimize model parameters; Personalized behavior correction solution generation module: Combined with generative adversarial network technology, the random noise vector input generator generates samples close to real driving behavior, while the discriminator distinguishes between real driving behavior samples and generated samples, and optimizes their respective parameters through adversarial training; Augmented reality (AR) feedback and interaction system: This system is installed on the head-up display HUD or AR glasses, processes the vehicle's sensor data in real time, and combines preset map data and traffic information to generate suggestions and warning information related to the current driving situation.

Images