A driving behavior recognition method, vehicle and computer-readable storage medium

Abstract

The present application provides a driving behavior identification method, a vehicle, and a computer-readable storage medium, the method comprising: obtaining multiple groups of driving behavior data during a user's driving; performing feature extraction on each group of driving behavior data to obtain multiple groups of driving behavior features. Sequentially identifying the driving states corresponding to the multiple groups of driving behavior features in chronological order, and obtaining multiple first driving states generated in chronological order during the user's driving. Performing cluster analysis on the multiple groups of driving behavior features to obtain multiple second driving states generated during the user's driving. A prediction result set is obtained by fusing the multiple groups of driving behavior features, the multiple first driving states, and the multiple second driving states. A driving behavior change list representing the changing rules of the driving behavior generated during the user's driving is obtained based on the prediction result set. This method can more accurately and comprehensively reflect the driving behavior of the user during driving.

Description

Technical Field

[0001] The present application relates to the field of vehicle control technology, and more specifically, to a driving behavior recognition method, a vehicle and a computer-readable storage medium in the field of vehicle control technology. Background Art

[0002] Understanding drivers' driving behavior is crucial to improving road safety, optimizing traffic management, enhancing driving experience, and formulating insurance policies for insurance companies. However, the existing driving behavior monitoring and analysis systems on the market cannot accurately identify complex driving behaviors, and have high false alarm and missed alarm rates when facing complex driving scenarios.

[0003] Therefore, it is urgent to propose effective solutions to the above problems. Summary of the invention

[0004] The present application provides a driving behavior recognition method, a vehicle, and a computer-readable storage medium. The present application can extract multiple sets of driving behavior features corresponding to multiple sets of driving behavior data during a user's driving period, and perform two driving state recognitions on the multiple sets of driving behavior features to determine multiple first driving states and multiple second driving states. The multiple first driving states can reflect the law of state changes during driving; the multiple second driving states include abnormal driving states not covered by the first driving states. By integrating the results of the two driving state recognitions and the driving behavior features during the user's driving period, the driving behavior of the user during driving can be more accurately and comprehensively reflected.

[0005] In a first aspect, a driving behavior recognition method is provided, the driving behavior recognition method comprising: obtaining multiple groups of driving behavior data generated by a vehicle during a user's driving; extracting driving behavior features from each group of driving behavior data to obtain multiple groups of driving behavior features; sequentially identifying driving states for the multiple groups of driving behavior features according to a time series to obtain multiple first driving states during the user's driving; wherein the multiple first driving states are arranged according to a time series; clustering the multiple groups of driving behavior features to obtain multiple second driving states during the user's driving; wherein the number of categories of the second driving states is greater than or equal to the number of categories of the first driving states; fusing the multiple groups of driving behavior features, the multiple first driving states, and the multiple second driving states to obtain a prediction result set of the user's driving behavior during driving; wherein the prediction result set includes probability values ​​of multiple driving behaviors; generating a driving behavior change list according to the prediction result set; wherein the driving behavior change list is used to represent the changing rules of normal driving behavior and / or abnormal driving behavior generated during the user's driving.

[0006] Through the above technical solution, multiple groups of driving behavior data generated by the vehicle during the user's driving are obtained; driving behavior features are extracted for each group of driving behavior data to obtain multiple groups of driving behavior features. The driving states corresponding to the multiple groups of driving behavior features are identified in chronological order, and multiple first driving states generated in chronological order during the user's driving are obtained. In addition, cluster analysis is performed on the multiple groups of driving behavior features to obtain multiple second driving states generated during the user's driving. The number of categories of the second driving state is greater than or equal to the number of categories of the first driving state. Then, by fusing the multiple groups of driving behavior features, the multiple first driving states and the multiple second driving states, a prediction result set of the driving behavior of the user during driving is obtained; the prediction result set includes probability values ​​of multiple driving behaviors. A driving behavior change list is generated according to the prediction result set. This application obtains multiple sets of driving behavior data during the user's driving, extracts multiple sets of driving behavior features corresponding to the multiple sets of driving behavior data, and performs two driving state recognitions on the multiple sets of driving behavior features. The first recognition will determine multiple first driving states, which can reflect the law of state changes during driving; the second recognition will determine multiple second driving states, which can identify abnormal driving states not covered by the first driving state. By integrating the results of the two driving state recognitions and the driving behavior features during the user's driving, the driving behavior of the user during driving can be more accurately and comprehensively reflected.

[0007] In combination with the first aspect, in some possible implementations, performing driving behavior feature extraction on each group of driving behavior data to obtain multiple groups of driving behavior features includes: using a convolutional layer of a pre-trained feature extraction model to extract local features of each group of driving behavior data to obtain multiple groups of local features, and performing a nonlinear transformation on the multiple groups of local features through an activation function to obtain multiple groups of first features; using a pooling layer of the feature extraction model to downsample the multiple groups of first features to obtain multiple groups of second features; wherein the dimension of the second feature is smaller than the dimension of the first feature; and using a fully connected layer of the feature extraction model to convert the multiple groups of first features and the multiple groups of second features into multiple groups of one-dimensional vectors of the same length, and performing a nonlinear transformation on the multiple groups of one-dimensional vectors of the same length through an activation function to obtain multiple groups of driving behavior features.

[0008] In combination with the first aspect and the above implementation, in some possible implementations, performing driving state recognition on multiple groups of driving behavior features in sequence according to a time series to obtain multiple first driving states during the user's driving includes: inputting multiple groups of driving behavior features into a pre-trained driving state recognition model; for each group of driving behavior features, based on the initial driving state distribution, the driving state transition probability matrix and the observation probability matrix, determining the probability of each driving state corresponding to the driving behavior feature to obtain multiple probability values; determining the driving state corresponding to the maximum value of the multiple probability values ​​as the target driving state of the driving behavior feature, and sequentially obtaining the target driving states corresponding to each of the multiple groups of driving behavior features; wherein the target driving states corresponding to each of the multiple groups of driving behavior features are arranged in a time series; and the target driving states corresponding to each of the multiple groups of driving behavior features are used as multiple first driving states. Based on the above technical solution, by performing driving state recognition on multiple groups of driving behavior features in sequence according to a time series, the driving states corresponding to each of the multiple groups of driving behavior features are identified, which can reflect the transformation rules of multiple driving states during the user's driving.

[0009] In combination with the first aspect and the above-mentioned implementation manner, in some possible implementation manners, clustering multiple groups of driving behavior features to obtain multiple second driving states during the user's driving includes: using a pre-trained driving state clustering model to initially cluster the multiple groups of driving behavior features to obtain multiple initial clusters; wherein the driving state category corresponding to each of the multiple initial clusters is the first driving state category; for a target cluster, determining the cluster center of the target cluster, obtaining the distance value between each group of driving behavior features in the target cluster and the cluster center, to obtain multiple distance values; wherein the target cluster is any one of the multiple initial clusters; determining the driving behavior features corresponding to the distance values ​​less than or equal to the distance threshold among the multiple distance values is the first feature, and the driving state corresponding to the first feature is determined as the first state to obtain multiple first states; wherein the first state is the driving state of the first driving state category corresponding to the target cluster; the driving behavior feature corresponding to the distance value greater than the distance threshold in the distance value is determined as the second feature; wherein the driving state category corresponding to the second feature is the second driving state category, and the first driving state category is different from the second driving state category; the driving state of the second driving state category corresponding to the second feature is determined as the second state to obtain multiple second states; multiple first states and multiple second states are determined as multiple second driving states; wherein the driving state category of multiple first driving states is the first driving state category. Based on the above technical solution, multiple groups of driving behavior features are clustered through the driving state recognition model, and abnormal states that cannot be identified by time series modeling of the driving behavior features are identified, and multiple second driving states are obtained, which can provide more detailed driving state classification results.

[0010] In combination with the first aspect and the above-mentioned implementation, in some possible implementations, fusing multiple sets of driving behavior features, multiple first driving states, and multiple second driving states to obtain a prediction result set of the user's driving behavior during driving includes: inputting multiple sets of driving behavior features, multiple first driving states, and multiple second driving states into a pre-trained fusion model; using the fusion model to obtain the weight values ​​corresponding to the feature extraction model, the driving state recognition model, and the driving state clustering model, respectively, to obtain multiple weight values; performing weighted operations on multiple sets of driving behavior features, multiple first driving states, and multiple second driving states based on the multiple weight values ​​to obtain a prediction result set. Based on the above technical solution, by fusing multiple sets of driving behavior features, multiple first driving states, and multiple second driving states to obtain a prediction result set, it is possible to accurately and comprehensively reflect whether the driving behavior corresponding to each time step is normal driving behavior or abnormal driving behavior, as well as the changing rules of normal driving behavior and / or abnormal driving behavior generated during the user's driving.

[0011] In combination with the first aspect and the above implementation, in some possible implementations, after generating a driving behavior change list according to the prediction result set, the driving behavior recognition method further includes: updating a model sample set according to multiple sets of driving behavior data, multiple sets of driving behavior features, multiple first driving states, multiple second driving states, and at least one of the prediction result sets; and optimizing the model parameters of at least one of the feature extraction model, the driving state recognition model, the driving state clustering model, and the fusion model using the updated model sample set. Based on the above technical solution, through this adaptive update, the model can be continuously optimized to adapt to changes in different driving scenarios and driving behaviors.

[0012] In combination with the first aspect and the above-mentioned implementation methods, in some possible implementation methods, generating a driving behavior change list based on the prediction result set includes: replacing the probability values ​​in the prediction result set that are greater than or equal to the probability threshold with a first identifier, and replacing the probability values ​​in the prediction result set that are less than the probability threshold with a second identifier, so as to generate the driving behavior change list; wherein the first identifier indicates that the driving behavior corresponding to the probability value is normal driving behavior, and the second identifier indicates that the driving behavior corresponding to the probability value is abnormal driving behavior.

[0013] In combination with the first aspect and the above-mentioned implementation methods, in some possible implementation methods, after generating a driving behavior change list based on the prediction result set, the driving behavior identification method also includes at least one of the following schemes: judging whether the user has committed insurance fraud based on the changing patterns of normal driving behavior and / or abnormal driving behavior generated during the user's driving; evaluating the risk level of the user's driving behavior based on the changing patterns of normal driving behavior and / or abnormal driving behavior generated during the user's driving.

[0014] In a second aspect, a driving behavior recognition device is provided, the driving behavior recognition device comprising:

[0015] An acquisition module, used to acquire multiple sets of driving behavior data generated by the vehicle during the user's driving;

[0016] A processing module is used to extract driving behavior features from each group of driving behavior data to obtain multiple groups of driving behavior features; and, sequentially identify driving states for the multiple groups of driving behavior features according to a time series to obtain multiple first driving states during a user's driving; wherein the multiple first driving states are arranged according to a time series; and, cluster the multiple groups of driving behavior features to obtain multiple second driving states during a user's driving; wherein the number of categories of the second driving states is greater than or equal to the number of categories of the first driving states; and, fuse the multiple groups of driving behavior features, the multiple first driving states, and the multiple second driving states to obtain a prediction result set of the user's driving behavior during driving; wherein the prediction result set includes probability values ​​of multiple driving behaviors; and, generate a driving behavior change list according to the prediction result set; wherein the driving behavior change list is used to represent the changing rules of normal driving behavior and / or abnormal driving behavior generated during a user's driving.

[0017] In a third aspect, a vehicle is provided, the vehicle comprising: a memory for storing executable program code;

[0018] A processor is used to call and run executable program code from a memory so that the vehicle executes the driving behavior recognition method in the above-mentioned first aspect or any possible implementation of the first aspect.

[0019] In a fourth aspect, a computer program product is provided, which includes: a computer program code, which, when executed on a computer, enables the computer to execute the driving behavior recognition method in the first aspect or any possible implementation of the first aspect.

[0020] In a fifth aspect, a computer-readable storage medium is provided, which stores a computer program code. When the computer program code runs on a computer, the computer executes the driving behavior recognition method in the above-mentioned first aspect or any possible implementation of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic flow chart of a driving behavior recognition method provided in an embodiment of the present application is shown;

[0022] Figure 2 A schematic diagram of the structure of a driving behavior recognition device provided in an embodiment of the present application is shown;

[0023] Figure 3 A schematic structural diagram of a vehicle provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] The technical solution in the present application will be described clearly and in detail below in conjunction with the accompanying drawings. In the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B: "and / or" in the text is only a description of the association relationship of associated objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, "multiple" means two or more than two.

[0025] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as suggesting or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features.

[0026] Understanding drivers' driving behavior is crucial to improving road safety, optimizing traffic management, enhancing driving experience, and formulating insurance policies for insurance companies. However, the existing driving behavior monitoring and analysis systems on the market cannot accurately identify complex driving behaviors, and have high false alarm and missed alarm rates when facing complex driving scenarios.

[0027] Based on the problems existing in the above-mentioned related technologies, the embodiments of the present application provide a driving behavior recognition method, a vehicle and a computer-readable storage medium. The present application obtains multiple sets of driving behavior data during the user's driving, extracts multiple sets of driving behavior features corresponding to the multiple sets of driving behavior data, and performs two driving state recognitions on the multiple sets of driving behavior features. The first recognition will determine multiple first driving states, and the multiple first driving states can reflect the law of state changes during the driving process; the second recognition will determine multiple second driving states, and the second recognition can identify abnormal driving states not covered by the first driving state. By integrating the results of the two driving state recognitions and the driving behavior features during the user's driving, the driving behavior of the user during driving can be more accurately and comprehensively reflected.

[0028] The following is an embodiment of a driving behavior recognition method provided in an embodiment of the present application.

[0029] Figure 1 A schematic flow chart of a driving behavior recognition method provided by an embodiment of the present application is shown, such as Figure 1 As shown, the embodiment of the present application provides a driving behavior recognition method. The driving behavior recognition method includes the following schemes:

[0030] S110: Acquire multiple sets of driving behavior data generated by the vehicle during the user's driving.

[0031] S120: Extracting driving behavior features from each set of driving behavior data to obtain multiple sets of driving behavior features.

[0032] S130: Perform driving state recognition on multiple groups of driving behavior features in sequence according to the time series to obtain multiple first driving states during the user's driving.

[0033] S140: Clustering the multiple groups of driving behavior features to obtain multiple second driving states during the user's driving.

[0034] S150: Fusing the multiple sets of driving behavior features, the multiple first driving states, and the multiple second driving states to obtain a prediction result set of the user's driving behavior during driving.

[0035] S160: Generate a driving behavior change list according to the prediction result set.

[0036] With the continuous advancement of intelligent connected vehicle technology, vehicle sensors and Internet of Vehicles devices can collect a large amount of driving behavior data about vehicles in real time. Among them, driving behavior data includes, for example, acceleration, steering angle, vehicle speed, and throttle opening. For example, the rotation speed of each tire can be measured by the vehicle's wheel speed sensor to calculate the vehicle speed, and the acceleration of the vehicle can be measured by the vehicle's acceleration sensor. Based on the vehicle's driving behavior data, car companies and insurance companies can deeply analyze users' driving behavior, assess risks, and identify potential abnormal operations or fraudulent behavior.

[0037] It should be understood that the driving behavior data is raw data directly obtained through vehicle sensors and has not been processed. Moreover, in the present application, multiple groups of driving behavior data are arranged in chronological order. For example, 5 groups of driving behavior data are obtained during the user's driving: the first group of driving behavior data-the fifth group of driving behavior data, wherein the acquisition time of the first group of driving behavior data-the fifth group of driving behavior data are 15:00, 15:01, 15:02, 15:03, and 15:04, respectively. Extracting driving behavior features from each group of driving behavior data refers to extracting features that can describe driving behavior from the raw data to facilitate subsequent data processing and analysis. After feature extraction, each group of driving behavior data will obtain one or more groups of driving behavior features. Among them, multiple groups of driving behavior features are also arranged in chronological order, and multiple groups of driving behavior features are represented in the form of vectors.

[0038] In driving behavior recognition, time series represents the order of time. Driving state recognition refers to analyzing each set of driving behavior features, identifying the driving state corresponding to each time step, and obtaining multiple first driving states. Among them, multiple first driving states are arranged in a time series, reflecting the changes in the user's driving state during driving. Among them, the categories of multiple first driving states may include pre-defined categories, such as "normal driving", "aggressive driving", "dangerous driving", etc. For example, multiple groups of driving behavior data include the first group of driving behavior data to the fifth group of driving behavior data, and the five first driving states obtained include: "normal driving", "aggressive driving", "dangerous driving", "normal driving" and "aggressive driving", and these five first driving states are arranged in chronological order.

[0039] It should be understood that there are some abnormal driving states that cannot be identified by time series modeling of driving behavior features. Therefore, in the embodiment of the present application, multiple groups of driving behavior features can be clustered to group similar driving behavior features to provide a more detailed driving state classification result. Specifically, the feature vectors are divided into different clusters by calculating the similarity between each group of driving behavior feature vectors and the reference feature vector. Among them, each cluster represents a category of driving state, the reference feature vector is the coordinate vector of the cluster center of each cluster, and the similarity can be reflected by the distance between the driving behavior feature vector and the reference feature vector. The multiple driving states obtained by cluster analysis are called the second driving state. For example, the number of clusters to be divided is preset, that is, the number of categories of driving states. For example, 3 categories are selected: "normal driving", "aggressive driving" and "dangerous driving", and 3 clusters are obtained. Secondly, the driving behavior features of the categories corresponding to each cluster are preset, and the preset driving behavior feature vector of each category is used as the reference feature vector. Assume that 10 groups of driving behavior features are included. First, the similarity between 10 groups of driving behavior feature vectors and the preset driving behavior feature vectors corresponding to the “normal driving” category is calculated, and it is determined that 2 groups of driving behavior features belong to the cluster corresponding to the “normal driving” category, that is, the driving state corresponding to the 2 groups of driving behavior features is “normal driving”; then, the similarity between the remaining 8 groups of driving behavior feature vectors and the preset driving behavior feature vectors corresponding to the “aggressive driving” category is calculated, and the driving state corresponding to 3 groups of driving behavior features is determined to be “aggressive driving”; finally, the similarity between the remaining 5 groups of driving behavior feature vectors and the preset driving behavior feature vectors corresponding to the “dangerous driving” category is calculated, and the driving state corresponding to 4 groups of driving behavior features is determined to be “dangerous driving”; then the driving state corresponding to the remaining 1 group of driving behavior features is “abnormal driving”.

[0040] It should be understood that driving state identification based on multiple groups of driving behavior features in sequence according to a time series cannot identify unknown abnormal driving behaviors, and cluster analysis can make up for this deficiency. Abnormal driving states that are not covered by the first driving state can be identified through cluster analysis. That is, the number of categories of the second driving state is greater than or equal to the number of categories of the first driving state. For example, the categories of multiple first driving states include "normal driving", "aggressive driving", and "dangerous driving", and the categories of multiple second driving states include "normal driving", "aggressive driving", "dangerous driving" and "abnormal driving". Among them, multiple second driving states are arranged in a time series, which more comprehensively reflects the user's driving state during driving.

[0041] Since, according to multiple sets of driving behavior characteristics and any one or any two of the multiple first driving states and the multiple second driving states, it is impossible to fully reflect the driving behavior of the user during driving. Therefore, in an embodiment of the present application, a prediction result set can be obtained by fusing multiple sets of driving behavior characteristics, multiple first driving states and multiple second driving states. Among them, the prediction result set includes probability values ​​of multiple driving behaviors. For example, the prediction result set A can be [0.2, 0.7, 0.8, 0.5, 0.4]. Among them, the prediction result set A includes probability values ​​of 5 driving behaviors corresponding to 5 time steps. A driving behavior change list can be generated according to the probability values ​​in the prediction result set; wherein, the driving behavior change list can fully reflect whether the driving behavior corresponding to each time step is normal driving behavior or abnormal driving behavior, as well as the changing rules of normal driving behavior and / or abnormal driving behavior generated during the user's driving.

[0042] Exemplarily, in a possible implementation, generating a driving behavior change list according to the prediction result set includes:

[0043] The probability values ​​in the prediction result set that are greater than or equal to the probability threshold are replaced with the first identifier, and the probability values ​​in the prediction result set that are less than the probability threshold are replaced with the second identifier to generate the driving behavior change list; wherein the first identifier indicates that the driving behavior corresponding to the probability value is normal driving behavior, and the second identifier indicates that the driving behavior corresponding to the probability value is abnormal driving behavior.

[0044] For example, assuming that the prediction result set A is [0.2, 0.7, 0.8, 0.5, 0.4], and the probability threshold is 0.7, the probability values ​​0.7 and 0.8 in the prediction result set A that are greater than or equal to the probability threshold are replaced with the first identifier a representing normal driving behavior, and the probability values ​​0.2, 0.5, and 0.4 in the prediction result set A that are less than the probability threshold are replaced with the second identifier b representing abnormal driving behavior. The converted prediction result set A is [b, a, a, b, b]. That is, the driving behavior change list is: abnormal driving behavior-normal driving behavior-normal driving behavior-abnormal driving behavior-abnormal driving behavior.

[0045] Through the above technical solution, multiple sets of driving behavior data during the user's driving are obtained, multiple sets of driving behavior features corresponding to the multiple sets of driving behavior data are extracted, and two driving state recognitions are performed on the multiple sets of driving behavior features. The first recognition will determine multiple first driving states, and the multiple first driving states can reflect the law of state changes during the driving process; the second recognition will determine multiple second driving states, and the second recognition can identify abnormal driving states not covered by the first driving state. By integrating the results of the two driving state recognitions and the driving behavior features during the user's driving, the driving behavior of the user during driving can be more accurately and comprehensively reflected.

[0046] In a possible implementation, the driving behavior feature extraction is performed on each set of driving behavior data to obtain multiple sets of driving behavior features including:

[0047] The convolution layer of the pre-trained feature extraction model is used to extract the local features of each set of driving behavior data to obtain multiple sets of local features, and the multiple sets of local features are nonlinearly transformed through an activation function to obtain multiple sets of first features;

[0048] Down-sampling the multiple groups of first features using a pooling layer of a feature extraction model to obtain multiple groups of second features;

[0049] The fully connected layer of the feature extraction model is used to convert multiple groups of first features and multiple groups of second features into multiple groups of one-dimensional vectors of the same length, and the multiple groups of one-dimensional vectors of the same length are nonlinearly transformed through an activation function to obtain multiple groups of driving behavior features.

[0050] Exemplarily, the feature extraction model in the embodiment of the present application can be a convolutional neural network (CNN). CNN is mainly composed of a convolution layer, a pooling layer, an activation function, a fully connected layer, etc. Among them, the convolution layer extracts local features by sliding on the input data by applying multiple convolution kernels. Each convolution kernel can capture different types of local features, such as edges, textures, etc. The output of the convolution layer is a set of feature maps. The feature map represents the local features of the input data. Exemplarily, the convolution operation formula can be expressed as:

[0051]

[0052] in, represents the input of the convolutional layer, is the convolution kernel weight of the convolution layer, * represents the convolution operation, is the bias term, is the activation function, is the output feature map of the convolutional layer.

[0053] Since the input of each convolutional layer is a linear function of the output of the upper convolutional layer, the entire convolutional neural network is a multi-layer linear transformation. The superposition of multiple layers of linear transformations is still a linear transformation. The ability of linear transformation is very limited and can only represent linear relationships, while many problems and data in application scenarios are nonlinear. Pure linear transformation cannot approximate complex nonlinear functions. Therefore, in an embodiment of the present application, a nonlinear function can be introduced as an activation function. Common nonlinear activation functions include ReLU (Rectified Linear Unit), Sigmoid, Tanh, etc. The function of the nonlinear activation function is to perform a nonlinear transformation on the input signal, so that the entire convolutional neural network has a stronger expressive ability. In addition, by introducing appropriate nonlinear activation functions (such as ReLU) in model training, the gradient vanishing problem can be effectively alleviated. Exemplarily, the activation function ReLU formula can be expressed as:

[0054]

[0055] Exemplarily, in the embodiment of the present application, the convolution layer is used to extract local features of each set of driving behavior data, and to perform nonlinear transformation of multiple sets of local features through an activation function to obtain multiple sets of first features. Among them, the local features of the driving behavior data are, for example, sudden acceleration, sudden braking, etc.

[0056] Since the first feature extracted by the convolution layer retains the local information of the input data, it has a high dimension and high computational complexity. Therefore, it needs to be input to the pooling layer for further processing.

[0057] The role of the pooling layer is to reduce the dimension of the feature map output by the convolutional layer by downsampling, thereby reducing the computational complexity and retaining the most important data features. Commonly used pooling operations include Max Pooling and Average Pooling.

[0058] Exemplarily, the maximum pooling formula can be expressed as:

[0059]

[0060] The maximum pooling layer is used to extract the maximum value in the feature map and retain the most significant features in the local area.

[0061] Exemplarily, in the embodiment of the present application, the pooling layer is used to downsample the multiple groups of first features, reduce the dimensions of the first features, and obtain multiple groups of second features, wherein the dimensions of the second features are smaller than the dimensions of the first features.

[0062] The function of the fully connected layer is to convert the feature maps extracted by the convolution layer and the pooling layer into the final output of the convolutional neural network. Exemplarily, in the embodiment of the present application, the fully connected layer is used to convert multiple groups of first features and multiple groups of second features into multiple groups of one-dimensional vectors of the same length, and perform nonlinear transformation on the multiple groups of one-dimensional vectors of the same length through an activation function to obtain multiple groups of driving behavior features.

[0063] It should be understood that before the next step of feature extraction is performed on multiple sets of driving behavior data, it is necessary to clean, denoise and normalize the multiple sets of driving behavior data. For example, first, interpolation or mean is used to fill missing values ​​to ensure the integrity of the driving behavior data; second, outliers are identified and deleted through statistical analysis to improve the quality of the driving behavior data; finally, the driving behavior data is formatted in chronological order to ensure that it can be used for time series modeling.

[0064] In a possible implementation, the driving state recognition is performed on multiple groups of driving behavior features in sequence according to the time series to obtain multiple first driving states during the user's driving, including:

[0065] Inputting multiple sets of driving behavior features into a pre-trained driving state recognition model;

[0066] For each set of driving behavior characteristics, based on the initial driving state distribution, the driving state transition probability matrix and the observation probability matrix, the probability of each driving state corresponding to the driving behavior characteristics is determined to obtain multiple probability values;

[0067] The driving state corresponding to the maximum value among the multiple probability values ​​is determined as the target driving state of the driving behavior feature, and the target driving states corresponding to the multiple groups of driving behavior features are obtained in turn; the target driving states corresponding to the multiple groups of driving behavior features are used as multiple first driving states.

[0068] It should be understood that the driving state recognition model in the embodiment of the present application can be a hidden Markov model (HMM). HMM is a statistical model suitable for processing time series data, such as driving behavior characteristics. HMM can be used to model the changing trend of driving behavior and identify different driving states. Specifically, the main parameters of HMM include initial state distribution, state transition matrix and observation probability matrix. Among them, the initial state distribution represents the probability of each hidden state appearing at the beginning of the sequence. For example, assuming there are three hidden states: "normal driving", "aggressive driving" and "dangerous driving", the initial state distribution can be expressed as: π=[π1, π2, π3]. Among them, π1 represents the probability of "normal driving" at the beginning of the sequence; π2 represents the probability of "aggressive driving" at the beginning of the sequence; π3 represents the probability of "dangerous driving" at the beginning of the sequence. For example: π=[0.6, 0.3, 0.1] This means that at the beginning of the sequence, the probability of "normal driving" is 0.6, the probability of "aggressive driving" is 0.3, and the probability of "dangerous driving" is 0.1.

[0069] The state transition matrix represents the probability of transitioning from one hidden state to another. For example, assuming there are three hidden states, the state transition matrix can be expressed as: A = . Where a_{ij} represents the probability of transitioning from state (i) to state (j). For example: A= . This means: the probability of going from "normal driving" to "normal driving" is 0.7, the probability of going to "aggressive driving" is 0.2, and the probability of going to "dangerous driving" is 0.1. The probability of going from "aggressive driving" to "normal driving" is 0.1, the probability of going to "aggressive driving" is 0.6, and the probability of going to "dangerous driving" is 0.3. The probability of going from "dangerous driving" to "normal driving" is 0.2, the probability of going to "aggressive driving" is 0.3, and the probability of going to "dangerous driving" is 0.5.

[0070] The observation probability matrix represents the probability of observing a specific observation under a certain hidden state. For example, suppose there are three hidden states and three observation values ​​(such as acceleration, braking force, steering angle). The observation probability matrix can be expressed as: B = Where: b_{ij} represents the probability of observing observation value (j) in state (i). For example: B= . This means: in the "normal driving" state, the probability of observing "acceleration" is 0.8, the probability of observing "braking force" is 0.1, and the probability of observing "steering angle" is 0.1. In the "aggressive driving" state, the probability of observing "acceleration" is 0.1, the probability of observing "braking force" is 0.7, and the probability of observing "steering angle" is 0.2. In the "dangerous driving" state, the probability of observing "acceleration" is 0.1, the probability of observing "braking force" is 0.1, and the probability of observing "steering angle" is 0.8.

[0071] In an embodiment of the present application, the target driving states corresponding to each of the multiple groups of driving behavior features are arranged in a time series. Exemplarily, for each group of driving behavior features, the most likely hidden state corresponding to each group of driving behavior features can be found through the Viterbi algorithm. Specifically, the probability of each driving state corresponding to the driving behavior feature is determined to obtain multiple probability values. The driving state corresponding to the maximum value of the multiple probability values ​​is determined as the target driving state of the driving behavior feature. For example, the probability of each driving state corresponding to a certain driving behavior feature is: normal driving 0.3, aggressive driving 0.5, and dangerous driving 0.8. Then the target driving state of the driving behavior feature is dangerous driving. The target driving states corresponding to each of the multiple groups of driving behavior features are obtained in sequence, and the target driving states corresponding to each of the multiple groups of driving behavior features are used as multiple first driving states. Exemplarily, multiple first driving states are, for example: normal driving, aggressive driving, normal driving, dangerous driving, and aggressive driving.

[0072] Based on the above technical solution, by sequentially identifying driving states for multiple groups of driving behavior features according to a time series, the driving states corresponding to the multiple groups of driving behavior features are identified, which can reflect the change rules of multiple driving states during the user's driving.

[0073] In a possible implementation, clustering the multiple groups of driving behavior features to obtain multiple second driving states during the user's driving includes:

[0074] Using a pre-trained driving state clustering model to perform initial clustering on multiple groups of driving behavior features to obtain multiple initial clusters; wherein the driving state category corresponding to each of the multiple initial clusters is a first driving state category;

[0075] For a target cluster, determine the cluster center of the target cluster, obtain the distance value between each group of driving behavior features in the target cluster and the cluster center, so as to obtain multiple distance values; wherein the target cluster is any one of the multiple initial clusters;

[0076] Determine the driving behavior feature corresponding to the distance value less than or equal to the distance threshold among the multiple distance values ​​as the first feature, and determine the driving state corresponding to the first feature as the first state, so as to obtain multiple first states; wherein the first state is a driving state of the first driving state category corresponding to the target cluster;

[0077] Determine the driving behavior feature corresponding to the distance value greater than the distance threshold among the distance values ​​as the second feature; wherein the driving state category corresponding to the second feature is the second driving state category, and the first driving state category is different from the second driving state category;

[0078] determining a driving state of a second driving state category corresponding to the second feature as a second state, so as to obtain a plurality of second states;

[0079] The plurality of first states and the plurality of second states are determined as a plurality of second driving states; wherein the driving state category of the plurality of first driving states is the first driving state category.

[0080] It should be understood that the driving state recognition model in the embodiment of the present application can be a K-Means clustering algorithm. The specific process of clustering multiple groups of driving behavior features using the K-Means clustering algorithm is as follows:

[0081] First, preset the number of clusters to be divided, that is, the number of categories of driving status. For example, select 3 categories: "normal driving", "aggressive driving" and "dangerous driving". Secondly, initialize the cluster center of each category. And by calculating the similarity between each set of driving behavior feature vectors and the reference feature vector, assign each set of driving behavior features to the category where the nearest cluster center is located. Repeat the steps of classifying the driving behavior features and updating the cluster center until the cluster center no longer changes or the maximum number of iterations is reached, and multiple initial clusters are obtained. Among them, each initial cluster corresponds to a category of driving status.

[0082] Then, the distance from each set of driving behavior features in each initial cluster to the cluster center of the initial cluster to which it belongs is calculated using the Euclidean distance formula to obtain multiple distance values. For example, the Euclidean distance formula can be expressed as:

[0083]

[0084] in,( , , ) is the coordinate of the driving behavior feature, ( , , ) are the coordinates of the cluster centers.

[0085] The driving behavior feature corresponding to the distance value less than or equal to the distance threshold among multiple distance values ​​is determined as the first feature, and the driving state corresponding to the first feature is determined as the first state, so as to obtain multiple first states. Among them, the first state is the driving state of the category corresponding to the initial cluster to which the group of driving behavior features belongs. For example, the category corresponding to the initial cluster to which the driving behavior feature a belongs is "normal driving", and the distance value from the driving behavior feature a to the cluster center of the category to which it belongs is less than the distance threshold, then the driving behavior feature a is determined as the first feature, and the driving state corresponding to the driving behavior feature a is determined as "normal driving". The driving behavior feature corresponding to the distance value greater than the distance threshold among multiple distance values ​​is determined as the second feature, that is, the abnormal behavior feature, and the driving state corresponding to the second feature is determined as the second state, that is, "abnormal driving". The multiple second states and the multiple first states are used as multiple second driving states. For example, the category corresponding to the initial cluster to which the driving behavior feature b belongs is "normal driving", and the distance value from the driving behavior feature b to the cluster center of the category to which it belongs is greater than the distance threshold, indicating that the driving behavior feature b does not belong to the initial cluster corresponding to the "normal driving" category. The driving behavior feature b is determined to be an abnormal behavior feature, and the driving state corresponding to the driving behavior feature b is determined to be "abnormal driving".

[0086] Based on the above technical solution, multiple groups of driving behavior features are clustered through the driving state recognition model to identify abnormal driving states that cannot be identified by time series modeling of driving behavior features, thereby providing more detailed driving state classification results.

[0087] In a possible implementation, the above-mentioned fusion of multiple sets of driving behavior features, multiple first driving states, and multiple second driving states to obtain a set of prediction results of the user's driving behavior during driving includes:

[0088] inputting a plurality of sets of driving behavior features, a plurality of first driving states, and a plurality of second driving states into a pre-trained fusion model;

[0089] A fusion model is used to obtain weight values ​​corresponding to the feature extraction model, the driving state recognition model, and the driving state clustering model, thereby obtaining a plurality of weight values;

[0090] A weighted operation is performed on multiple groups of driving behavior features, multiple first driving states, and multiple second driving states based on multiple weight values ​​to obtain a prediction result set.

[0091] Exemplarily, the fusion model is, for example, logistic regression or random forest, and the fusion formula can be expressed as:

[0092]

[0093] in, are the outputs of the feature extraction model, the driving state recognition model, and the driving state clustering model, respectively. is the model weight. For example, Driving behavior characteristics, The first driving state, For the second driving state, is the driving state clustering model weight. g is the activation function.

[0094] Based on the above technical solution, by combining the results of the two driving state recognitions and introducing a comprehensive analysis of the driving behavior characteristics of the user during driving, a prediction result set is obtained, which can accurately and comprehensively reflect the driving behavior of the user during driving.

[0095] In a possible implementation, after generating the driving behavior change list according to the prediction result set, the driving behavior recognition method further includes:

[0096] updating the model sample set according to at least one of the plurality of sets of driving behavior data, the plurality of sets of driving behavior features, the plurality of first driving states, the plurality of second driving states, and the prediction result set;

[0097] The updated model sample set is used to optimize model parameters of at least one of the feature extraction model, the driving state recognition model, the driving state clustering model, and the fusion model.

[0098] This application designs an online learning mechanism that can update model parameters in real time through new data. The new data includes: multiple sets of driving behavior data, multiple sets of driving behavior features, multiple first driving states and multiple second driving states, and at least one of the prediction result sets. Specifically, after each driving behavior recognition is completed and the driving behavior change list is generated, the new data will be added to the training set and the model will be updated using an incremental learning algorithm. The online learning formula is as follows:

[0099]

[0100] in, is the parameter set of all current models, is the learning rate, is the loss function, For new data, is the change in the loss function, is a constant.

[0101] Through this adaptive update, the model can be continuously optimized to adapt to changes in different driving scenarios and driving behaviors.

[0102] In a possible implementation, after generating the driving behavior change list according to the prediction result set, the driving behavior recognition method further includes at least one of the following solutions:

[0103] Determine whether the user has committed insurance fraud based on the changing patterns of the user's normal driving behavior and / or abnormal driving behavior during driving;

[0104] The risk level of the user's driving behavior is assessed based on the changing patterns of the normal driving behavior and / or abnormal driving behavior generated during the user's driving.

[0105] Exemplarily, the changing patterns of normal driving behavior and / or abnormal driving behavior generated by the user during driving include the following uses. Use 1: Based on the changing patterns of normal driving behavior and / or abnormal driving behavior of the user during driving, the insurance company can determine whether the user has committed insurance fraud and reasonably formulate insurance policies. For example, after user A is involved in a traffic accident, if the insurance company finds that user A has multiple consecutive abnormal driving behaviors (such as speeding, sudden braking, etc.) during the damage assessment, then the insurance company has reason to suspect that user A may have committed insurance fraud, and the insurance company can consider these factors in the division of liability. Therefore, the insurance company can reduce the amount of compensation based on this information to ensure that the claims process is more fair and reasonable.

[0106] Purpose 2: Based on the changing rules of normal driving behavior and / or abnormal driving behavior during driving, the risk level of the user's driving behavior can be evaluated to remind the user to drive in a standardized manner and improve driving safety. For example, once abnormal behavior (such as sudden acceleration, sudden braking, etc.) is found, the system will promptly feedback to the driver through the vehicle display, voice or mobile application to reduce the potential risk of accidents. For example, assuming that user B has the following abnormal driving behaviors (speeding, sudden braking, frequent lane changes, etc.) many times during driving within a certain period of time, the system will assess user B's driving behavior as a high-risk level based on the frequency and severity of these abnormal behaviors. Subsequently, a warning message is sent to user B through the vehicle display, mobile phone application or text message, reminding him to pay attention to driving safety and suggesting improvements to specific driving behaviors. In addition, if user B applies for a driver's job, the employer can use user B's driving behavior risk level as reference information to decide whether to hire user B. This helps employers more accurately assess the driving ability and safety of candidates, thereby making more informed recruitment decisions.

[0107] Through the above technical solution, multiple sets of driving behavior data during the user's driving are obtained, multiple sets of driving behavior features corresponding to the multiple sets of driving behavior data are extracted, and two driving state recognitions are performed on the multiple sets of driving behavior features. The first recognition will determine multiple first driving states, and the multiple first driving states can reflect the law of state changes during the driving process; the second recognition will determine multiple second driving states, and the second recognition can identify abnormal driving states not covered by the first driving state. By integrating the results of the two driving state recognitions and the driving behavior features during the user's driving, the driving behavior of the user during driving can be more accurately and comprehensively reflected.

[0108] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the driving behavior recognition device of the present application, please refer to the embodiment of the driving behavior recognition method of the present application.

[0109] Figure 2 The schematic diagram of the structure of a driving behavior recognition device provided in an embodiment of the present application is shown. The driving behavior recognition device 200 includes:

[0110] An acquisition module 210 is used to acquire multiple sets of driving behavior data generated by the vehicle during the user's driving;

[0111] The processing module 220 is used to extract driving behavior features from each group of driving behavior data to obtain multiple groups of driving behavior features; and, identify driving states for the multiple groups of driving behavior features in sequence according to a time series to obtain multiple first driving states during the user's driving; wherein the multiple first driving states are arranged according to a time series; and, cluster the multiple groups of driving behavior features to obtain multiple second driving states during the user's driving; wherein the number of categories of the second driving states is greater than or equal to the number of categories of the first driving states; and, fuse the multiple groups of driving behavior features, the multiple first driving states, and the multiple second driving states to obtain a prediction result set of the user's driving behavior during driving; wherein the prediction result set includes probability values ​​of multiple driving behaviors; and, generate a driving behavior change list according to the prediction result set; wherein the driving behavior change list is used to represent the changing rules of normal driving behavior and / or abnormal driving behavior generated during the user's driving.

[0112] In a possible implementation, the processing module 220 is specifically configured to:

[0113] The convolution layer of the pre-trained feature extraction model is used to extract the local features of each set of driving behavior data to obtain multiple sets of local features, and the multiple sets of local features are nonlinearly transformed through an activation function to obtain multiple sets of first features;

[0114] Downsampling the multiple groups of first features using a pooling layer of a feature extraction model to obtain multiple groups of second features; wherein the dimension of the second features is smaller than the dimension of the first features;

[0115] The fully connected layer of the feature extraction model is used to convert multiple groups of first features and multiple groups of second features into multiple groups of one-dimensional vectors of the same length, and the multiple groups of one-dimensional vectors of the same length are nonlinearly transformed through an activation function to obtain multiple groups of driving behavior features.

[0116] In a possible implementation, the processing module 220 is specifically configured to:

[0117] Inputting multiple sets of driving behavior features into a pre-trained driving state recognition model;

[0118] For each set of driving behavior characteristics, based on the initial driving state distribution, the driving state transition probability matrix and the observation probability matrix, the probability of each driving state corresponding to the driving behavior characteristics is determined to obtain multiple probability values;

[0119] The driving state corresponding to the maximum value among the multiple probability values ​​is determined as the target driving state of the driving behavior feature, and the target driving states corresponding to the multiple groups of driving behavior features are obtained in sequence; wherein the target driving states corresponding to the multiple groups of driving behavior features are arranged in time series;

[0120] The target driving states corresponding to the multiple groups of driving behavior characteristics are used as multiple first driving states.

[0121] In a possible implementation, the processing module 220 is specifically configured to:

[0122] Using a pre-trained driving state clustering model to perform initial clustering on multiple groups of driving behavior features to obtain multiple initial clusters; wherein the driving state category corresponding to each of the multiple initial clusters is a first driving state category;

[0123] For a target cluster, determine the cluster center of the target cluster, obtain the distance value between each group of driving behavior features in the target cluster and the cluster center, so as to obtain multiple distance values; wherein the target cluster is any one of the multiple initial clusters;

[0124] Determine the driving behavior feature corresponding to the distance value less than or equal to the distance threshold among the multiple distance values ​​as the first feature, and determine the driving state corresponding to the first feature as the first state, so as to obtain multiple first states; wherein the first state is a driving state of the first driving state category corresponding to the target cluster;

[0125] Determine the driving behavior feature corresponding to the distance value greater than the distance threshold among the distance values ​​as the second feature; wherein the driving state category corresponding to the second feature is the second driving state category, and the first driving state category is different from the second driving state category;

[0126] determining a driving state of a second driving state category corresponding to the second feature as a second state, so as to obtain a plurality of second states;

[0127] The plurality of first states and the plurality of second states are determined as a plurality of second driving states; wherein the driving state category of the plurality of first driving states is the first driving state category.

[0128] In a possible implementation, the processing module 220 is specifically configured to:

[0129] inputting a plurality of sets of driving behavior features, a plurality of first driving states, and a plurality of second driving states into a pre-trained fusion model;

[0130] A fusion model is used to obtain weight values ​​corresponding to the feature extraction model, the driving state recognition model, and the driving state clustering model, thereby obtaining a plurality of weight values;

[0131] A weighted operation is performed on multiple groups of driving behavior features, multiple first driving states, and multiple second driving states based on multiple weight values ​​to obtain a prediction result set.

[0132] In a possible implementation, the processing module 220 is further configured to:

[0133] updating the model sample set according to at least one of the plurality of sets of driving behavior data, the plurality of sets of driving behavior features, the plurality of first driving states, the plurality of second driving states, and the prediction result set;

[0134] The updated model sample set is used to optimize model parameters of at least one of the feature extraction model, the driving state recognition model, the driving state clustering model, and the fusion model.

[0135] In a possible implementation, the processing module 220 is specifically configured to:

[0136] The probability values ​​in the prediction result set that are greater than or equal to the probability threshold are replaced with the first identifier, and the probability values ​​in the prediction result set that are less than the probability threshold are replaced with the second identifier to generate the driving behavior change list; wherein the first identifier indicates that the driving behavior corresponding to the probability value is normal driving behavior, and the second identifier indicates that the driving behavior corresponding to the probability value is abnormal driving behavior.

[0137] In a possible implementation, the processing module 220 is further configured to:

[0138] Determine whether the user has committed insurance fraud based on the changing patterns of the user's normal driving behavior and / or abnormal driving behavior during driving;

[0139] The risk level of the user's driving behavior is assessed based on the changing patterns of the normal driving behavior and / or abnormal driving behavior generated during the user's driving.

[0140] Figure 3 It is a structural schematic diagram of a vehicle provided in an embodiment of the present application.

[0141] For example, Figure 3 As shown, the vehicle 300 includes: a memory 301 and a processor 302, wherein the memory 301 stores an executable program code 3011, and the processor 302 is used to call and execute the executable program code 3011 to perform a driving behavior recognition method.

[0142] In this embodiment, the functional modules of the vehicle can be divided according to the above method example. For example, each functional module can be corresponded, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware. It should be noted that the division of modules in this embodiment is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0143] In the case of dividing each functional module according to each function, the vehicle may include: an acquisition module, a processing module, a control module, etc. It should be noted that all relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module, which will not be repeated here.

[0144] The vehicle provided in this embodiment is used to execute the above-mentioned driving behavior recognition method, so the same effect as the above-mentioned implementation method can be achieved.

[0145] In the case of an integrated unit, the vehicle may include a processing module and a storage module. The processing module may be used to control and manage the actions of the vehicle. The storage module may be used to support the vehicle in executing relevant program codes and data.

[0146] The processing module may be a processor or a controller, which may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. The processor may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc. The storage module may be a memory.

[0147] This embodiment also provides a computer-readable storage medium, in which a computer program code is stored. When the computer program code is executed on a computer, the computer executes the above-mentioned related method steps to implement a driving behavior recognition method in the above-mentioned embodiment.

[0148] This embodiment also provides a computer program product. When the computer program product is run on a computer, the computer is caused to execute the above-mentioned related steps to implement a driving behavior recognition method in the above-mentioned embodiment.

[0149] In addition, the vehicle provided in the embodiments of the present application may specifically be a chip, component or module, and the vehicle may include a connected processor and memory; wherein the memory is used to store instructions, and when the vehicle is running, the processor may call and execute instructions so that the chip executes a driving behavior recognition method in the above-mentioned embodiments.

[0150] The vehicle, computer-readable storage medium, computer program product or chip provided in this embodiment are all used to execute the corresponding driving behavior recognition method provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding driving behavior recognition method provided above, and will not be repeated here.

[0151] Through the description of the above implementation methods, technical personnel in the relevant field can 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.

[0152] In the embodiments provided in 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 only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0153] The above contents are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A driving behavior recognition method, characterized in that: The driving behavior recognition method comprises: Acquire multiple sets of driving behavior data generated by the vehicle during the user's driving; Extracting driving behavior features from each set of driving behavior data to obtain multiple sets of driving behavior features; Performing driving state recognition on the multiple groups of driving behavior features in sequence according to the time series to obtain multiple first driving states during the user's driving; wherein the multiple first driving states are arranged in a time series; Clustering the multiple groups of driving behavior features to obtain multiple second driving states during the user's driving; wherein the number of categories of the second driving states is greater than or equal to the number of categories of the first driving state; The plurality of driving behavior feature sets, the plurality of first driving states, and the plurality of second driving states are integrated to obtain a prediction result set of the driving behavior of the user during driving; wherein the prediction result set includes probability values ​​of the plurality of driving behaviors; Generate a driving behavior change list according to the prediction result set; wherein the driving behavior change list is used to indicate the change pattern of normal driving behavior and / or abnormal driving behavior generated during the user's driving; The clustering of the plurality of groups of driving behavior features to obtain a plurality of second driving states during the user's driving comprises: Performing initial clustering on the multiple groups of driving behavior features using a pre-trained driving state clustering model to obtain multiple initial clusters; wherein the driving state category corresponding to each of the multiple initial clusters is a first driving state category; For a target cluster, determining a cluster center of the target cluster, obtaining a distance value between each group of driving behavior features in the target cluster and the cluster center to obtain a plurality of distance values; wherein the target cluster is any one of the plurality of initial clusters; Determine the driving behavior feature corresponding to the distance value less than or equal to the distance threshold among the multiple distance values ​​as a first feature, and determine the driving state corresponding to the first feature as a first state, so as to obtain multiple first states; wherein the first state is a driving state of a first driving state category corresponding to the target cluster; Determine the driving behavior feature corresponding to the distance value greater than the distance threshold among the multiple distance values ​​as a second feature; wherein the driving state category corresponding to the second feature is a second driving state category, and the first driving state category is different from the second driving state category; determining a driving state of a second driving state category corresponding to the second feature as a second state, so as to obtain a plurality of second states; The multiple first states and the multiple second states are determined as the multiple second driving states; wherein the driving state category of the multiple first driving states is the first driving state category.

2. The driving behavior recognition method according to claim 1, characterized in that: The driving behavior feature extraction is performed on each set of driving behavior data to obtain multiple sets of driving behavior features, including: Using a convolutional layer of a pre-trained feature extraction model to extract local features of each set of driving behavior data to obtain multiple sets of local features, and performing nonlinear transformation on the multiple sets of local features through an activation function to obtain multiple sets of first features; Down-sampling the multiple groups of first features using the pooling layer of the feature extraction model to obtain multiple groups of second features; wherein the dimension of the second feature is smaller than the dimension of the first feature; The multiple groups of first features and the multiple groups of second features are converted into multiple groups of one-dimensional vectors of the same length using the fully connected layer of the feature extraction model, and the multiple groups of one-dimensional vectors of the same length are nonlinearly transformed through the activation function to obtain the multiple groups of driving behavior features.

3. The driving behavior recognition method according to claim 2, characterized in that: The step of sequentially identifying the driving states of the plurality of groups of driving behavior features according to the time series to obtain a plurality of first driving states during the user's driving comprises: inputting the plurality of sets of driving behavior features into a pre-trained driving state recognition model; For each set of driving behavior features, based on the initial driving state distribution, the driving state transition probability matrix and the observation probability matrix, the probability of each driving state corresponding to the driving behavior features is determined to obtain multiple probability values; Determine the driving state corresponding to the maximum value among the multiple probability values ​​as the target driving state of the driving behavior feature, and sequentially obtain the target driving states corresponding to each of the multiple groups of driving behavior features; wherein the target driving states corresponding to each of the multiple groups of driving behavior features are arranged in time series; The target driving states corresponding to the multiple groups of driving behavior features are used as the multiple first driving states.

4. The driving behavior recognition method according to claim 1, characterized in that: The step of fusing the multiple sets of driving behavior features, the multiple first driving states, and the multiple second driving states to obtain a prediction result set of the driving behavior of the user during driving includes: inputting the plurality of groups of driving behavior features, the plurality of first driving states, and the plurality of second driving states into a pre-trained fusion model; Using the fusion model to obtain weight values ​​corresponding to the feature extraction model, the driving state recognition model, and the driving state clustering model, respectively, to obtain a plurality of weight values; The plurality of groups of driving behavior features, the plurality of first driving states, and the plurality of second driving states are weightedly calculated based on the plurality of weight values ​​to obtain the prediction result set.

5. The driving behavior recognition method according to claim 4, characterized in that: After generating a driving behavior change list according to the prediction result set, the driving behavior identification method further includes: updating the model sample set according to the multiple sets of driving behavior data, the multiple sets of driving behavior features, the multiple first driving states, the multiple second driving states, and at least one of the prediction result sets; The updated model sample set is used to optimize model parameters of at least one of the feature extraction model, the driving state recognition model, the driving state clustering model, and the fusion model.

6. The driving behavior recognition method according to any one of claims 1 to 5, characterized in that: Generating a driving behavior change list according to the prediction result set includes: The probability values ​​in the prediction result set that are greater than or equal to the probability threshold are replaced with the first identifier, and the probability values ​​in the prediction result set that are less than the probability threshold are replaced with the second identifier to generate the driving behavior change list; wherein the first identifier indicates that the driving behavior corresponding to the probability value is normal driving behavior, and the second identifier indicates that the driving behavior corresponding to the probability value is abnormal driving behavior.

7. The driving behavior recognition method according to any one of claims 1 to 5, characterized in that: After generating a driving behavior change list according to the prediction result set, the driving behavior identification method further includes at least one of the following solutions: Determining whether the user has committed insurance fraud based on the changing patterns of the user's normal driving behavior and / or abnormal driving behavior during driving; The risk level of the user's driving behavior is evaluated according to the changing pattern of the normal driving behavior and / or abnormal driving behavior generated during the user's driving.

8. A vehicle, characterized in that: The vehicle comprises: A memory for storing executable program codes; A processor, configured to call and run the executable program code from the memory, so that the vehicle executes the driving behavior recognition method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the driving behavior recognition method according to any one of claims 1 to 7 is implemented.

Images