Method and device for allocating human-machine co-driving control rights considering driving group characteristics

Abstract

The present application relates to the field of data processing technology, and discloses a method and device for allocating human-machine co-driving control rights that takes into account the characteristics of a driving group. The method includes: acquiring vehicle operation data and extracting behavioral features, constructing a driving group identification model based on deep neural networks and dual-path features; classifying driver behaviors to generate group labels, and obtaining control parameters based on the labels; establishing a dual-exponential control rights allocation model through lateral deviation and steering torque, and combining dynamic weights based on group identification frequency to obtain allocation results. The present application improves the efficiency and accuracy of allocating human-machine co-driving control rights that takes into account the characteristics of a driving group.

Description

Technical Field

[0001] The present application relates to the field of data processing, and more particularly to a method and device for allocating human-machine co-driving control rights taking into account the characteristics of a driving group. Background Art

[0002] In the prior art, the human-machine co-driving control rights allocation method mainly relies on preset fixed allocation rules, or performs simple weight calculations based on vehicle status and driver operating intentions. These methods usually adopt a single control strategy, such as a linear allocation method based on vehicle lateral deviation, a proportional allocation method based on driver steering torque, etc. At the same time, some methods introduce intelligent algorithms such as fuzzy control or neural networks to try to improve the flexibility of control rights allocation. However, most of these methods do not take into account the differences among driving groups, and ignore the significant differences between different drivers in terms of operating habits, reaction characteristics, and driving skills.

[0003] The existing technology has the following deficiencies: first, the control right allocation strategy is too rigid and cannot adapt to the operating preferences of different driving groups, resulting in frequent control rights disputes during human-machine collaboration; second, there is a lack of in-depth analysis and real-time recognition mechanism of driver behavior characteristics, and it is impossible to accurately judge the type of driving group to which the current driver belongs; third, the dynamic changes of driving group characteristics over time are not fully considered in the control right allocation process, making it difficult to achieve a smooth transition of control rights; finally, when dealing with complex driving scenarios, the existing methods often have the problem of unreasonable control right allocation, which affects driving safety and comfort. Summary of the invention

[0004] The present application provides a method and device for allocating human-machine co-driving control rights considering the characteristics of a driving group, which is used to improve the efficiency and accuracy of allocating human-machine co-driving control rights considering the characteristics of a driving group.

[0005] In a first aspect, the present application provides a method for allocating human-machine co-driving control rights considering the characteristics of a driving group, and the method for allocating human-machine co-driving control rights considering the characteristics of a driving group includes:

[0006] Collecting vehicle operation data through on-board sensors, extracting following vehicle distance data, lateral displacement data, vehicle acceleration data, and driving operation frequency data from the vehicle operation data to obtain driving behavior characteristic data;

[0007] A deep neural network model is established using the driving behavior feature data, and a driving group recognition model is constructed based on a dual-path architecture of manually extracted features and CNN convolutional features;

[0008] Classifying and distinguishing the real-time behaviors of drivers according to the driving group identification model to generate general driving group labels, aggressive driving group labels and novice driving group labels;

[0009] Acquire corresponding control parameters according to the driving group label, and calculate the vehicle lateral comprehensive deviation based on the normalized weighted combination of the preview point lateral deviation and the yaw angle deviation;

[0010] By using the vehicle lateral comprehensive deviation and the driver's steering torque, a control right allocation mathematical model is established through a double exponential function to obtain an intelligent system control weight coefficient and a driver control weight coefficient;

[0011] According to the driving group label and the group identification frequency data within the time window, the intelligent system control weight coefficient and the driver control weight coefficient are dynamically weighted and combined to generate a control right allocation result.

[0012] In a second aspect, the present application provides a human-machine co-driving control rights allocation device considering the characteristics of a driving group, and the human-machine co-driving control rights allocation device considering the characteristics of a driving group includes:

[0013] An extraction module is used to collect vehicle operation data through vehicle-mounted sensors, extract vehicle-to-vehicle distance data, lateral displacement data, vehicle acceleration data, and driving operation frequency data from the vehicle operation data, and obtain driving behavior characteristic data;

[0014] A construction module is used to establish a deep neural network model using the driving behavior feature data, and to construct a driving group recognition model based on a dual-path architecture of manually extracted features and CNN convolutional features;

[0015] A discrimination module, used to classify and discriminate the real-time behaviors of drivers according to the driving group recognition model, and generate general driving group labels, aggressive driving group labels and novice driving group labels;

[0016] A combination module, used for obtaining corresponding control parameters according to the driving group label, and calculating the vehicle lateral comprehensive deviation based on the normalized weighted combination of the preview point lateral deviation and the yaw angle deviation;

[0017] A distribution module, used to establish a control right distribution mathematical model by using the vehicle lateral comprehensive deviation and the driver's steering torque through a double exponential function to obtain an intelligent system control weight coefficient and a driver control weight coefficient;

[0018] The weighting module is used to dynamically weight the intelligent system control weight coefficient and the driver control weight coefficient according to the driving group label and the group identification frequency data in the time window to generate a control right allocation result.

[0019] In the technical solution provided by the present application, the present invention collects vehicle operation data through on-board sensors and extracts vehicle-to-vehicle distance data, lateral displacement data, vehicle acceleration data and driving operation frequency data, thereby providing rich data support for the comprehensive analysis of driving behavior characteristics and ensuring the accurate capture of driver operation characteristics; a deep neural network model is established using driving behavior feature data, and a dual-path architecture of manual feature extraction and CNN convolutional features is adopted to improve the accuracy and robustness of feature extraction and enhance the model's ability to recognize different driving behavior patterns; the real-time behavior of drivers is classified and judged through a driving group recognition model, and general driving group labels, aggressive driving group labels and novice driving group labels are generated, thereby realizing the recognition of driver groups. Accurate identification and dynamic tracking of body characteristics; based on the driving group label, the corresponding control parameters are obtained, and the normalized weighted combination of the lateral deviation of the preview point and the yaw angle deviation is combined to achieve the accurate calculation of the vehicle's lateral comprehensive deviation, providing a reliable basis for the allocation of control rights; the vehicle's lateral comprehensive deviation and the driver's steering torque are used to construct a mathematical model of control rights allocation of a double exponential function, and the intelligent system control weight coefficient and the driver's control weight coefficient are obtained, ensuring the smoothness and continuity of the control rights allocation; the control weight coefficient is dynamically weighted and combined according to the driving group label and the group recognition frequency data within the time window, so that the control rights allocation result can adapt to the dynamic changes of the driver's behavior characteristics, and improve the adaptability and coordination of the human-machine co-driving system. Through the organic combination of the above technical features, the present invention not only improves the accuracy of driving group identification, but also realizes the adaptive allocation of control rights, effectively reduces the control rights conflict in the process of human-machine co-driving, while ensuring driving safety and comfort, and providing personalized human-machine co-driving experience for different driving groups. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0021] Figure 1 A schematic diagram of an embodiment of a method for allocating human-machine co-driving control rights taking into account driving group characteristics in an embodiment of the present application;

[0022] Figure 2 This is a schematic diagram of an embodiment of a device for allocating human-machine co-driving control rights taking into account the characteristics of a driving group in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The embodiments of the present application provide a method and device for allocating human-machine co-driving control rights taking into account the characteristics of the driving group. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the method for allocating human-machine co-driving control rights considering the characteristics of the driving group includes:

[0025] Step S101, collecting vehicle operation data through vehicle-mounted sensors, extracting vehicle-to-vehicle distance data, lateral displacement data, vehicle acceleration data, and driving operation frequency data from the vehicle operation data, and obtaining driving behavior characteristic data;

[0026] Step S102: using the driving behavior feature data to establish a deep neural network model, and constructing a driving group recognition model based on a dual-path architecture of manually extracted features and CNN convolutional features;

[0027] Step S103: classify and identify the real-time behaviors of the drivers according to the driving group recognition model, and generate general driving group labels, aggressive driving group labels, and novice driving group labels;

[0028] Step S104, obtaining corresponding control parameters according to the driving group label, and calculating the vehicle lateral comprehensive deviation based on the normalized weighted combination of the preview point lateral deviation and the yaw angle deviation;

[0029] Step S105, using the vehicle's lateral comprehensive deviation and the driver's steering torque, a control right allocation mathematical model is established through a double exponential function to obtain an intelligent system control weight coefficient and a driver control weight coefficient;

[0030] Step S106: dynamically weight the intelligent system control weight coefficient and the driver control weight coefficient according to the driving group label and the group identification frequency data in the time window to generate a control right allocation result.

[0031] It is understandable that the execution subject of the present application can be a human-machine co-driving control rights allocation device that considers the characteristics of the driving group, or a terminal or a server, which is not limited here. The embodiment of the present application is described by taking the server as the execution subject as an example.

[0032] Specifically, the raw data during the operation of the vehicle is collected through the on-board sensor system. The on-board sensor system includes millimeter-wave radar, visual sensor and CAN bus. The millimeter-wave radar is mainly used to collect the distance information of the vehicle in front, with a detection range of 0-200 meters and a sampling frequency of 20Hz; the visual sensor collects lane line information for calculating lateral displacement, with a field of view of 120 degrees and a resolution of 1920×1080 pixels; the CAN bus collects the vehicle's own acceleration, speed and other state information, with a sampling frequency of 100Hz. The raw data is processed by a multi-sensor data fusion algorithm. The fusion algorithm uses the Kalman filter method to align and fuse the data collected by different sensors in time and space dimensions to generate a unified data format. After data fusion, the following vehicle distance data is obtained with a data resolution of 0.1 meters; the lateral displacement data has an accuracy of centimeters; and the vehicle acceleration data has an accuracy of 0.01m / s². Based on these basic data, the driving operation frequency is calculated through a 30-second sliding time window, including the number of sudden braking, sudden acceleration and sharp turning, where sudden braking is defined as a deceleration greater than 4m / s², sudden acceleration is defined as an acceleration greater than 3m / s², and sharp turning is defined as a lateral acceleration greater than 2m / s². After obtaining the driving behavior feature data, a deep neural network model is established to construct a driving group recognition model. The model adopts a dual-path architecture, namely the manual feature extraction path and the CNN convolution feature path. In the manual feature extraction path, the driving behavior feature data is statistically extracted, including statistics such as mean, standard deviation, kurtosis, and skewness, to form a 128-dimensional feature vector. The CNN convolution path uses a three-layer convolution structure with convolution kernel sizes of 3×3, 5×5, and 7×7, respectively. Each layer is followed by a maximum pooling layer, and finally a 256-dimensional feature vector is obtained through a fully connected layer. After the two feature vectors are connected in series, they pass through two fully connected layers and finally access the softmax classification layer to obtain a driving group recognition model.

[0033] According to the trained driving group recognition model, the real-time behavior of drivers is classified and judged. The model outputs the probability values ​​of three categories, corresponding to the general driving group, the aggressive driving group and the novice driving group. The probability threshold of 0.6 is used for discrimination. When the probability value of a category exceeds the threshold, it is judged as that category. To avoid frequent switching, a 5-second time window is used for smoothing, and the category with the highest frequency in the window is taken as the final group label. After obtaining the driving group label, the control parameters are set according to the characteristics of different groups. Taking the preview distance of 30 meters as an example, the lateral deviation at the preview point is calculated, and the vehicle yaw angle deviation is obtained at the same time. The two deviation values ​​are normalized and mapped to the [0,1] interval. Different driving groups use different weighting coefficients. The lateral deviation weight of the general driving group is 0.6, and the yaw angle deviation weight is 0.4; the aggressive driving group is 0.7 and 0.3 respectively; the novice driving group is 0.5 and 0.5 respectively. The vehicle lateral comprehensive deviation is obtained by weighted combination.

[0034] Based on the vehicle's lateral comprehensive deviation and the driver's steering torque, a mathematical model for control rights allocation is constructed. The model uses a double exponential function structure to establish exponential mappings for the comprehensive deviation and steering torque respectively. Different driving groups correspond to different exponential function parameters: the general driving group parameters are =4, =3, =0.5; the aggressive driving group parameter is =4, =10, =0.5; the parameter of the novice driving group is =20, =3, β3=0.5. The control weight coefficients of the intelligent system and the driver are calculated through these parameters. Finally, the frequency data of group identification is statistically analyzed using a time window with a window length of 10 seconds. The dynamic weight coefficient is calculated based on the frequency data, and the control weight coefficients of the intelligent system and the driver are weighted and combined. For example, within a 10-second window, the driver's behavior is identified as the general driving group 6 times and the aggressive driving group 4 times. The weight of the general driving group is 0.6, and the weight of the aggressive driving group is 0.4. These weights are multiplied by their respective control weight coefficients and summed to obtain the final control right allocation result.

[0035] For example, when the driver is driving on the highway, the vehicle is running at a speed of 100 km / h. The collected data shows that the distance from the vehicle in front fluctuates around 50 meters, the average lateral deviation is 0.2 meters, the acceleration changes smoothly, and one sharp turn operation is recorded within 30 seconds. After the driving group identification model is used for discrimination, the output probability distribution is: 0.75 for the general driving group, 0.15 for the aggressive driving group, and 0.1 for the novice driving group, which is determined to be the general driving group. At this time, the lateral deviation at the preview point is 0.3 meters, and the yaw angle deviation is 2 degrees. After normalization, they are 0.3 and 0.2 respectively, and the weighted comprehensive deviation is 0.26. The steering torque applied by the driver is 2N·m. Through the control right allocation model, the intelligent system control weight is 0.3 and the driver control weight is 0.7. In the subsequent 10-second window, the identification result of the general driving group is maintained, and the final control right allocation result remains at this level.

[0036] It should be noted that, in the embodiment of the present application, a control rights allocation model for human-machine co-driving is designed taking into account the driver's steering behavior and the lateral deviation of the vehicle.

[0037]

[0038] In the formula, , Respectively represent the real-time control rights of the intelligent system and the driver. , , is a parameter related to the characteristics of the driving group, represents the normalized steering torque, represents the normalized vehicle lateral comprehensive deviation. Among them, , are the normalized values ​​of the lateral deviation and yaw angle deviation at the preview point, respectively. The design parameters and satisfy .

[0039] Design corresponding driving group characteristic parameters for different driving groups.

[0040] (1) General driving group, characteristic parameters meet

[0041] (2) Aggressive driving group, characteristic parameters meet

[0042] (3) Novice driver group, characteristic parameters meet

[0043] For the control rights allocation model of the general driving group, as the comprehensive deviation As the size of the intelligent system increases, the control of the intelligent system will gradually increase, and as the driver gradually participates in the steering task ( )hour, The growth rate is relatively reduced. For the control rights allocation model of the aggressive driving group, the comprehensive deviation When smaller (such as ), the driver has always been in control, only when the comprehensive deviation The intelligent system intervenes only when a certain threshold is exceeded, which is more in line with the driving style of aggressive drivers. For the control allocation model of novice drivers, the intelligent system intervenes relatively early and quickly, and can compensate for the lateral control ability of novice drivers in a timely manner, thereby improving driving safety.

[0044] Taking into account the evolution of drivers' driving behavior habits, the driving group label to which they belong will change dynamically. Therefore, a flexible control rights allocation strategy is designed.

[0045] (1) The default label of the driving group is the general driving group;

[0046] (2) Based on the identification results of the driving group, a weighted combination method is used to achieve flexible allocation of human-machine control rights, that is,

[0047] When the real-time identification result is a general driving group,

[0048]

[0049] When the real-time identification result is an aggressive driving group,

[0050]

[0051] In the formula, , is the dynamic weight coefficient, which is determined by the frequency of identification results in a time sliding window of length L, where L is a design parameter.

[0052] When the real-time identification result is a novice driver group,

[0053]

[0054] In the formula, , is the dynamic weight coefficient, which is determined by the frequency of the identification results in a time sliding window of length L.

[0055] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0056] (1) Obtaining raw driving data of the vehicle during driving through millimeter-wave radar and visual sensors, and extracting basic driving parameters from the raw driving data;

[0057] (2) Perform multi-sensor data fusion processing on basic driving parameters to generate real-time following vehicle distance data, lateral displacement data, and vehicle acceleration data;

[0058] (3) Based on the real-time following vehicle distance data and lateral displacement data, the number of sudden braking, sudden acceleration and sharp turning are calculated through the sliding time window algorithm to obtain the driving operation frequency data;

[0059] (4) Using vehicle acceleration data and driving operation frequency data, a data normalization algorithm is used to perform feature normalization processing to obtain standardized driving characteristics;

[0060] (5) Preprocessing the standardized driving features through the CAN bus to generate a preprocessing feature sequence;

[0061] (6) The preprocessed feature sequence is subjected to feature combination mapping operation to output driving behavior feature data.

[0062] Specifically, the original driving data of the vehicle during driving is obtained through millimeter-wave radar and visual sensor. The millimeter-wave radar operates in the 77GHz frequency band, with a detection range of 0-200 meters, an angular resolution of 0.1 degrees, and a sampling frequency of 20Hz. It mainly collects information such as the distance and relative speed of the vehicle in front; the visual sensor uses a high-definition camera with a resolution of 1920×1080, a frame rate of 30fps, and a field of view of 120 degrees. It mainly collects lane line information, surrounding vehicle positions and other data. The collected original driving data is preliminarily processed to extract basic driving parameters, including the original value of the distance to the front vehicle, the original value of the lane deviation, and the original value of the speed. For the basic driving parameters obtained from multiple sensors, a multi-sensor data fusion processing method is used for data integration. The Kalman filter algorithm is used for data fusion to perform time alignment and spatial registration on the data collected by different sensors. The linear interpolation method is used for time alignment to unify the data with different sampling frequencies to a sampling frequency of 100Hz; the spatial registration converts the data in different coordinate systems into a unified vehicle body coordinate system through the sensor calibration matrix. After fusion processing, high-precision real-time vehicle-to-vehicle distance data is generated with an accuracy of 0.1 meter; lateral displacement data with an accuracy of centimeters; and vehicle acceleration data with an accuracy of 0.01m / s².

[0063] Based on the fused real-time following vehicle distance data and lateral displacement data, the sliding time window algorithm is used to calculate the driving operation frequency. The sliding window length is set to 30 seconds, the sliding step is 1 second, and the number of emergency brakes, emergency accelerations, and sharp turns are counted in each window. The emergency braking judgment standard is that the deceleration is greater than 4m / s² and the duration exceeds 0.5 seconds; the emergency acceleration judgment standard is that the acceleration is greater than 3m / s² and the duration exceeds 0.5 seconds; the sharp turn judgment standard is that the lateral acceleration is greater than 2m / s² and the duration exceeds 0.3 seconds. By detecting and counting these mutation behaviors, the driving operation frequency data is obtained. The vehicle acceleration data and driving operation frequency data are normalized. The z-score normalization method is used to calculate the mean and standard deviation of each feature, and the data is converted to a distribution with a mean of 0 and a standard deviation of 1. For the acceleration data, the statistical features of the longitudinal acceleration and lateral acceleration are calculated respectively; for the driving operation frequency data, the number of each type of operation within the 30-second window is converted into a frequency value per unit time. The standardization process ensures that the features of different dimensions are comparable and obtains standardized driving features.

[0064] The standardized driving characteristics are preprocessed through the CAN bus. The CAN bus uses a transmission rate of 500kbps to collect vehicle status information such as vehicle speed, steering wheel angle, and accelerator pedal opening. The collected data is noise filtered, and the median filter algorithm is used to eliminate outliers. The filter window size is 5 sampling points. Then the data is downsampled to reduce the data sampling rate to 10Hz, while retaining the key feature information of the data to generate a preprocessed feature sequence. Finally, the preprocessed feature sequence is subjected to feature combination mapping operation. The principal component analysis method is used to extract the main feature components and retain the feature dimensions with a cumulative contribution rate of 95%. Then, the feature selection algorithm is used to calculate the importance score of each feature, and the features with higher scores are selected to form the final feature set, and the driving behavior feature data is output.

[0065] For example, on a road with heavy traffic, the distance to the vehicle ahead fluctuates between 45 and 65 meters, with an average distance of 55 meters, and the visual sensor detects that the lateral deviation of the vehicle relative to the center line of the lane varies between -0.3 and 0.3 meters. After multi-sensor fusion processing, a stable following vehicle distance data sequence is obtained, eliminating the measurement noise and uncertainty of a single sensor. In a 30-second sliding window, 2 emergency braking operations (maximum decelerations of 4.5m / s² and 4.2m / s², respectively), 1 emergency acceleration operation (maximum acceleration of 3.3m / s²), and 1 sharp turn operation (maximum lateral acceleration of 2.3m / s²) are detected. These raw data are standardized, and the obtained eigenvalues ​​are distributed in the range of [-3,3]. The vehicle speed collected through the CAN bus is maintained between 80 and 90km / h, and the steering wheel angle changes within a range of ±45 degrees. After the feature combination mapping operation, the driving behavior feature data finally generated contains 15 key dimensions, which fully describes the driving behavior characteristics during this period.

[0066] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0067] (1) Input the driving behavior feature data into the pre-trained artificial feature extraction layer to generate an artificial feature vector;

[0068] (2) Extract the driving behavior feature data through the CNN convolution layer to obtain the CNN feature vector;

[0069] (3) Combine the artificial feature vector and the CNN feature vector in series to obtain a fused feature vector;

[0070] (4) Perform deep neural network training based on the fused feature vector to obtain deep feature representation;

[0071] (5) Use deep feature representation and prior knowledge to perform knowledge transfer learning to form knowledge-enhanced features;

[0072] (6) The knowledge-enhanced features are classified and learned through a multi-layer perceptron to build a driving group recognition model.

[0073] Specifically, the driving behavior feature data is input into the pre-trained artificial feature extraction layer. The artificial feature extraction layer contains multiple statistical feature extraction units to calculate the mean, standard deviation, kurtosis, skewness and other statistics of the driving behavior data. For the following distance data, the average following distance, distance fluctuation range, distance change rate and other features are extracted; for the lateral displacement data, the average offset, maximum offset, offset frequency and other features are extracted; for the acceleration data, the acceleration distribution characteristics, acceleration change rate and other information are extracted. After artificial feature extraction, a 128-dimensional artificial feature vector is generated. At the same time, the CNN convolutional network is used to perform deep feature extraction on the driving behavior feature data. The CNN network structure contains three convolutional layers. The first layer uses 32 3×3 convolution kernels with a step size of 1 and an output feature map size of 32×32×32; the second layer uses 64 5×5 convolution kernels with a step size of 2 and an output feature map size of 16×16×64; the third layer uses 128 7×7 convolution kernels with a step size of 2 and an output feature map size of 8×8×128. Each convolutional layer is followed by a ReLU activation function and a maximum pooling layer with a pooling kernel size of 2×2 and a step size of 2. Finally, the feature map is converted into a 256-dimensional CNN feature vector through a fully connected layer.

[0074] The 128-dimensional artificial feature vector and the 256-dimensional CNN feature vector are combined in series to form a 384-dimensional fused feature vector. The concatenation operation retains the advantages of the two feature extraction methods: the artificial feature extraction layer provides interpretable statistical features, and the CNN feature extraction layer captures the deep spatiotemporal features of the data. The fused feature vector contains comprehensive information about driving behavior and provides rich feature representation for subsequent deep learning. A deep neural network is constructed based on the fused feature vector for training. The network structure contains three fully connected layers with 256, 128 and 64 nodes respectively. The first and second layers use the ReLU activation function, and the Dropout layer is added to prevent overfitting, with a dropout rate of 0.5. During the training process, a mini-batch training method with a batch size of 32 is adopted, the Adam optimizer is used, the learning rate is set to 0.001, and the number of training rounds is 100. The network parameters are optimized by the back-propagation algorithm, and a 64-dimensional deep feature representation is finally obtained.

[0075] Deep feature representation is used for knowledge transfer learning, and prior knowledge in the field of driving behavior is introduced. Prior knowledge includes typical feature patterns of different driving groups, such as the stability feature of the general driving group, the quick response feature of the aggressive driving group, and the cautious feature of the novice driving group. Through the transfer learning method, these prior knowledge are integrated with the deep feature representation to enhance the discriminative ability of the feature. In the specific process, the knowledge distillation technology is used, and the pre-trained model is used as the teacher network to guide the feature learning process to form a 96-dimensional knowledge-enhanced feature. Finally, the knowledge-enhanced feature is input into the multi-layer perceptron for classification learning. The multi-layer perceptron contains two hidden layers with 64 and 32 nodes respectively, and uses the ReLU activation function. The output layer uses the softmax activation function to output the probability values ​​of three categories, corresponding to the general driving group, the aggressive driving group, and the novice driving group. In the training stage, the cross entropy loss function is used to optimize the network parameters through the back propagation algorithm. The training data set contains 5,000 labeled samples, with 20% of the validation set and 10% of the test set. After training, the classification accuracy of the model on the test set reaches 92%.

[0076] For example, the input driving behavior feature data contains 15 dimensions. After being processed by the artificial feature extraction layer, the calculated statistical features show that the average following distance is 52 meters, the standard deviation of the distance is 5.3 meters, the mean of the lateral offset is 0.15 meters, and the acceleration kurtosis is 2.8, generating a 128-dimensional artificial feature vector. At the same time, the CNN network performs convolution processing on the input data. The first layer of convolution extracts basic edge features, the second layer of convolution captures the speed change pattern, and the third layer of convolution identifies the complete driving behavior fragment, and finally generates a 256-dimensional CNN feature vector. After the two feature vectors are connected in series and trained by the deep neural network, the deep feature representation obtained clearly reflects the driver's operating characteristics. After combining prior knowledge for transfer learning, the knowledge enhancement feature better highlights the group characteristics. The classification results of the multi-layer perceptron show that the probability of the driver belonging to the general driving group is 0.85, the probability of belonging to the aggressive driving group is 0.1, and the probability of belonging to the novice driving group is 0.05. Finally, it is classified as the general driving group.

[0077] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0078] (1) Extract features from real-time driving behavior data through a driving group recognition model to generate a real-time feature vector;

[0079] (2) Calculate the score probability distribution based on the real-time feature vector to form the group category probability value;

[0080] (3) Use the group category probability value to set the threshold for multi-classification discrimination and output the initial group label;

[0081] (4) Perform time series smoothing on the initial group labels to obtain smoothed group labels;

[0082] (5) Perform group category mapping based on smoothed group labels to obtain standardized group labels;

[0083] (6) The standardized group labels are processed through label fusion to generate general driving group labels, aggressive driving group labels, and novice driving group labels.

[0084] Specifically, the real-time driving behavior data is feature extracted through the pre-trained driving group recognition model. The process adopts a dual-path feature extraction architecture. The manual feature extraction path performs statistical analysis on the driving data, including calculating the mean, standard deviation, maximum and minimum values ​​of the following distance, the range and frequency of lateral deviation, and the distribution characteristics of acceleration. The CNN feature extraction path extracts temporal features through a three-layer convolutional network, with convolution kernel sizes of 3×3, 5×5, and 7×7, respectively, and each layer is followed by a maximum pooling operation. The features extracted by the dual paths are combined in series to generate a 384-dimensional real-time feature vector. The score probability distribution is calculated based on the real-time feature vector, and the softmax function is used to map the features into probability values ​​of three driving group categories. The softmax function converts the output of the deep neural network into a probability value in the interval [0,1], and the sum of the probabilities of the three categories is 1. The probability calculation takes into account the weight contribution of each dimension in the feature vector, and the weight matrix of the fully connected layer is used for weighted combination to form the group category probability value.

[0085] The calculated probability values ​​of group categories are thresholded for multi-classification discrimination, and the threshold is set to 0.6. When the probability value of a certain category exceeds the threshold, the driver is judged as the corresponding driving group. If the probability values ​​of all categories do not exceed the threshold, the category with the highest probability is selected as the judgment result. The initial group label is obtained by threshold discrimination. In order to reduce the control instability caused by frequent label switching, the initial group label is time-series smoothed. A sliding time window of 5 seconds is used to perform statistical analysis on the label sequence in the window. The smoothing process uses the majority voting method to select the label with the most occurrences in the window as the label at the current moment. At the same time, the state transition constraint is introduced, requiring that the label changes at adjacent moments must meet certain smoothness conditions to obtain a smooth group label.

[0086] Based on the smooth group label, group category mapping is performed to convert discrete label values ​​into standardized representations. The mapping process takes into account the similarities and differences between different driving groups, and uses one-hot encoding to convert group labels into three-dimensional vector representations. The general driving group is mapped to [1,0,0], the aggressive driving group is mapped to [0,1,0], and the novice driving group is mapped to [0,0,1] to obtain standardized group labels. The standardized group labels are subjected to label fusion processing to generate the final driving group labels. The fusion process takes into account temporal continuity and spatial consistency, and uses the exponential weighted average method to accumulate historical labels. The weight coefficient decays exponentially over time, and the label at the most recent moment has a higher weight. Through fusion processing, stable general driving group labels, aggressive driving group labels, and novice driving group labels are obtained.

[0087] For example, during a driver's driving, real-time driving behavior data showed that the average following distance was maintained at 55 meters, the standard deviation was 4.8 meters, the lateral offset peak was 0.25 meters, and one sudden acceleration operation was recorded within 30 seconds. After feature extraction, a 384-dimensional feature vector was obtained, and the probability values ​​of the three categories were obtained through softmax calculation: 0.75 for the general driving group, 0.15 for the aggressive driving group, and 0.10 for the novice driving group. Since the probability of the general driving group exceeded the threshold of 0.6, the initial label was determined to be the general driving group. Within the 5-second time window, the results of 5 consecutive judgments were all general driving groups, and the labels were kept stable after smoothing. The standardized mapping obtained a vector representation of [1,0,0]. The final label fusion result confirmed that the driver belonged to the general driving group, providing a reliable group feature basis for the subsequent allocation of control rights.

[0088] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0089] (1) Extract the preview point location data from the vehicle status information to obtain the real-time coordinates of the preview point;

[0090] (2) Calculate the original value of the lateral deviation based on the real-time coordinates of the preview point to generate the lateral deviation of the preview point;

[0091] (3) Calculate the original value of the yaw angle using the vehicle posture data to obtain the yaw angle deviation;

[0092] (4) Normalizing the lateral deviation and yaw angle deviation of the preview point to obtain a normalized deviation value;

[0093] (5) Select the corresponding weighting coefficient according to the driving group label to generate the deviation weighting parameter;

[0094] (6) The normalized deviation value and the deviation weighting parameter are combined and calculated to obtain the vehicle lateral comprehensive deviation.

[0095] Specifically, the preview point position data is extracted from the vehicle status information. The preview point is set 30 meters in front of the vehicle, and the vehicle's global position coordinates (x, y) and heading angle θ are obtained through the on-board sensors. Based on the vehicle's current position and preview distance, the real-time coordinates (xp, yp) of the preview point are calculated using coordinate transformation. The coordinate transformation takes into account the influence of the vehicle's heading angle to ensure that the preview point is always in the direction of the vehicle's forward movement. Based on the real-time coordinates of the preview point, the lateral deviation is calculated. First, the reference trajectory equation of the lane centerline is obtained, and the shape of the lane line is described using a third-order polynomial curve fitting. The shortest distance from the preview point to the reference trajectory is calculated, which is the original value of the lateral deviation. In the specific calculation, the iterative nearest point algorithm is used to find the point on the reference trajectory closest to the preview point. The vertical distance between the two points is the lateral deviation of the preview point.

[0096] The yaw angle deviation is calculated using the vehicle attitude data. The vehicle attitude data includes information such as heading angle and yaw rate, which are collected by the inertial measurement unit with a sampling frequency of 100Hz. The angle between the vehicle heading angle and the tangent direction of the reference trajectory is calculated to obtain the original value of the yaw angle. Considering the dynamic characteristics of the vehicle, the yaw rate term is introduced for compensation to improve the accuracy of the angle estimation, and finally the yaw angle deviation is obtained. The lateral deviation and yaw angle deviation of the preview point are normalized. The normalized range of the lateral deviation is [-1,1], and the corresponding actual deviation range is [-2m,2m]; the normalized range of the yaw angle deviation is also [-1,1], and the corresponding actual angle range is [-10°,10°]. The linear normalization method is used to map the original value to a unified numerical range, which is convenient for the subsequent weighted combination calculation to obtain the normalized deviation value.

[0097] According to the characteristics of different driving groups, the corresponding weighting coefficients are selected. For the general driving group, the lateral deviation weight is 0.6 and the yaw angle deviation weight is 0.4, which reflects the moderate requirement for lane keeping; for the aggressive driving group, the lateral deviation weight is 0.7 and the yaw angle deviation weight is 0.3, emphasizing greater operational freedom; for the novice driving group, the lateral deviation weight is 0.5 and the yaw angle deviation weight is 0.5, reflecting a more conservative control strategy. According to the current driving group label, the corresponding weight coefficient combination is selected to generate the deviation weighting parameter. The normalized deviation value and the deviation weighting parameter are combined and calculated, and the weighted summation method is used to obtain the vehicle's lateral comprehensive deviation. The comprehensive deviation value is also normalized to the [0,1] interval. The larger the value, the farther it deviates from the expected driving state, and the stronger the control intervention is required.

[0098] For example: A vehicle is driving on a curve at a speed of 80 km / h. The vehicle's current position coordinates (100.5m, 50.3m) and heading angle of 45 degrees are obtained through the global positioning system. The coordinates of the preview point 30 meters ahead are calculated (121.7m, 71.5m). The reference trajectory equation provided by the lane recognition system shows that the original value of the lateral deviation at the preview point is 0.8 meters. The vehicle posture data shows that the current yaw angle is 3 degrees and the yaw rate is 0.2rad / s. After normalization, the normalized value of the lateral deviation is 0.4 and the normalized value of the yaw angle deviation is 0.3. The current driving group is identified as a general driving group. Using weight coefficients of 0.6 and 0.4, the calculated lateral comprehensive deviation is 0.36, indicating that the vehicle is in a controllable deviation state and requires moderate control intervention to maintain the lane.

[0099] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0100] (1) Establishing the first exponential function based on the vehicle's lateral comprehensive deviation to generate a deviation weight factor;

[0101] (2) Using the driver's steering torque to establish a second exponential function, the steering weight factor is obtained;

[0102] (3) Combining the deviation weight factor and the steering weight factor into a function to obtain a control weight reference value;

[0103] (4) Calculate the initial weight of the intelligent system according to the control weight reference value to obtain the control weight coefficient of the intelligent system;

[0104] (5) Calculate the initial weight of the driver based on the complementary relationship between the intelligent system and the driver’s weight;

[0105] (6) The driver’s initial weight is normalized to obtain the driver’s control weight coefficient.

[0106] Specifically, in the process of allocating control rights of human-machine co-driving considering the characteristics of the driving group, the first exponential function is first established according to the lateral comprehensive deviation of the vehicle. The exponential function uses the lateral comprehensive deviation as the independent variable and generates the deviation weight factor through exponential mapping. Different exponential parameters are used for different driving groups: the general driving group parameter is 4, the aggressive driving group parameter is 4, and the novice driving group parameter is 20. The numerical range of the deviation weight factor is limited to between 0 and 1, reflecting the degree of influence of the lateral deviation on the allocation of control rights. The second exponential function is established using the driver's steering torque. The function uses the steering torque as the input variable and obtains the steering weight factor through exponential operation. Different exponential parameters are set for different driving groups: the general driving group parameter is 3, the aggressive driving group parameter is 10, and the novice driving group parameter is 3. The steering weight factor is also limited to the range of 0 to 1, reflecting the strength of the driver's steering intention.

[0107] The deviation weight factor and the steering weight factor are combined as functions, and the combined function is constructed in the form of product. During the combination process, the deviation weight factor reflects the degree of deviation of the vehicle, and the steering weight factor reflects the driver's operating intention. The product of the two forms the control weight reference value, which realizes the coupling of deviation and steering intention. The initial weight of the intelligent system is calculated according to the control weight reference value. The adjustment coefficient 0.5 is introduced into the calculation process to adjust the reference value to ensure that the control weight of the intelligent system is maintained within a reasonable range. The adjusted value is the control weight coefficient of the intelligent system, which is used for subsequent control right allocation.

[0108] Based on the complementary relationship between the intelligent system and the driver's weight, the driver's initial weight is calculated. According to the constraint that the sum of control rights is 1, the driver's initial weight is equal to 1 minus the intelligent system control weight coefficient. This calculation method ensures a smooth distribution of control rights between humans and machines. The driver's initial weight is normalized and the weight value is mapped to the range of 0 to 1. The normalization process takes into account the maximum and minimum values ​​of the historical weights, and the standardized driver control weight coefficient is obtained through linear mapping.

[0109] For example, when the vehicle is driving on a curved road, the measured lateral comprehensive deviation is 0.36, the steering torque applied by the driver is 2 Nm, and the normalized value of the steering torque is 0.4. The current driving group is identified as a general driving group, and the deviation weight factor is calculated to be 0.23, and the steering weight factor is 0.70. The control weight baseline value after function combination is 0.161, and the intelligent system control weight coefficient is further calculated to be 0.098, and the driver's initial weight is 0.902. Finally, after normalization, the driver's control weight coefficient is 0.85, indicating that the driver maintains a dominant position in the current scenario and the intelligent system plays an auxiliary role.

[0110] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0111] (1) Count the group recognition results based on the time sliding window and generate group recognition frequency data;

[0112] (2) Calculate the dynamic weight coefficient using group identification frequency data to obtain the group weight parameter;

[0113] (3) Weighted combination of the intelligent system control weight coefficient and the group weight parameter to obtain the final weight of the intelligent system;

[0114] (4) Perform a weighted combination of the driver control weight coefficient and the group weight parameter to generate the driver's final weight;

[0115] (5) Calculate the control rights allocation based on the final weight of the intelligent system and the final weight of the driver;

[0116] (6) Normalize the control rights allocation calculation results to obtain the control rights allocation results.

[0117] Specifically, a time sliding window is used to count the group recognition results. The sliding window length is set to 10 seconds, the sliding step is 1 second, and the number of recognitions of the general driving group, the aggressive driving group, and the novice driving group is recorded in each window. The frequency of occurrence of each group is obtained by counting statistics to form group recognition frequency data. The dynamic weight coefficient is calculated based on the group recognition frequency data. The calculation process takes into account the proportion of each driving group in the window and converts the frequency data into weight parameters. For driving groups with higher frequency of occurrence, a larger weight value is assigned; for driving groups with lower frequency of occurrence, a smaller weight value is assigned. The group weight parameter is obtained through the mapping relationship from frequency to weight.

[0118] The control weight coefficient of the intelligent system is weighted combined with the group weight parameter. During the combination process, the control characteristics of different driving groups are considered, and the control weight of the intelligent system is dynamically adjusted. When the recognition result tends to be a novice driving group, the weight of the intelligent system is increased; when the recognition result tends to be an aggressive driving group, the weight of the intelligent system is reduced, thereby obtaining the final weight of the intelligent system. A similar weighted combination is performed on the driver control weight coefficient and the group weight parameter. When combining, the influence of the driving group characteristics on the driver's control right is considered. When identified as an aggressive driving group, the driver's control weight is increased; when identified as a novice driving group, the driver's control weight is appropriately reduced, and finally the driver's final weight is generated.

[0119] The control right allocation calculation is performed based on the final weight of the intelligent system and the final weight of the driver. The calculation process ensures that the sum of the two weights is 1, and at the same time considers the smoothness of the weight change to avoid incoherent operations caused by sudden changes in control rights. The control right allocation calculation results are normalized to ensure that the final control right allocation results are within a reasonable range. The normalization process uses a linear mapping method to map the weight value to a standard interval to obtain a standardized control right allocation result.

[0120] For example, within a 10-second sliding window, the driving group identification results show: 6 times for the general driving group, 3 times for the aggressive driving group, and 1 time for the novice driving group. The group weight parameters are calculated based on the frequency data: 0.6 for the general driving group, 0.3 for the aggressive driving group, and 0.1 for the novice driving group. The original control weight coefficient of the intelligent system is 0.3, and the final weight after weighted combination is 0.25; the original control weight coefficient of the driver is 0.7, and the final weight after weighted combination is 0.75. The control right allocation result after normalization remains unchanged, reflecting the control mode of driver-led and intelligent system-assisted in the current driving scenario.

[0121] The above describes the method for allocating human-machine co-driving control rights considering the characteristics of the driving group in the embodiment of the present application. The following describes the device for allocating human-machine co-driving control rights considering the characteristics of the driving group in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of a device for allocating human-machine co-driving control rights taking into account the characteristics of a driving group includes:

[0122] The extraction module 201 is used to collect vehicle operation data through vehicle-mounted sensors, extract vehicle-to-vehicle distance data, lateral displacement data, vehicle acceleration data and driving operation frequency data from the vehicle operation data, and obtain driving behavior characteristic data;

[0123] A construction module 202 is used to establish a deep neural network model using the driving behavior feature data, and to construct a driving group recognition model based on a dual-path architecture of manually extracted features and CNN convolutional features;

[0124] A discrimination module 203 is used to classify and discriminate the real-time behaviors of drivers according to the driving group recognition model, and generate general driving group labels, aggressive driving group labels and novice driving group labels;

[0125] A combination module 204 is used to obtain corresponding control parameters according to the driving group label, and calculate the vehicle lateral comprehensive deviation based on the normalized weighted combination of the preview point lateral deviation and the yaw angle deviation;

[0126] An allocation module 205 is used to establish a control right allocation mathematical model by using the vehicle lateral comprehensive deviation and the driver's steering torque through a double exponential function to obtain an intelligent system control weight coefficient and a driver control weight coefficient;

[0127] The weighting module 206 is used to dynamically weight the intelligent system control weight coefficient and the driver control weight coefficient according to the driving group label and the group identification frequency data in the time window to generate a control right allocation result.

[0128] Through the coordinated cooperation of the above-mentioned components, the present invention collects vehicle operation data through on-board sensors and extracts vehicle-to-vehicle distance data, lateral displacement data, vehicle acceleration data and driving operation frequency data, which provides rich data support for the comprehensive analysis of driving behavior characteristics and ensures the accurate capture of driver operation characteristics; uses driving behavior characteristic data to establish a deep neural network model, adopts a dual-path architecture of manual feature extraction and CNN convolutional features, improves the accuracy and robustness of feature extraction, and enhances the model's recognition ability for different driving behavior patterns; uses a driving group recognition model to classify and distinguish drivers' real-time behaviors, generates general driving group labels, aggressive driving group labels and novice driving group labels, and realizes the recognition of driving Accurate identification and dynamic tracking of group characteristics of drivers; based on the driving group label, the corresponding control parameters are obtained, and the normalized weighted combination of the lateral deviation of the preview point and the yaw angle deviation is combined to achieve the accurate calculation of the vehicle's lateral comprehensive deviation, providing a reliable basis for the allocation of control rights; the vehicle's lateral comprehensive deviation and the driver's steering torque are used to construct a mathematical model of control rights allocation of a double exponential function, and the intelligent system control weight coefficient and the driver's control weight coefficient are obtained, ensuring the smoothness and continuity of the control rights allocation; the control weight coefficient is dynamically weighted and combined according to the driving group label and the group recognition frequency data within the time window, so that the control rights allocation result can adapt to the dynamic changes of the driver's behavior characteristics, and improve the adaptability and coordination of the human-machine co-driving system. Through the organic combination of the above technical features, the present invention not only improves the accuracy of driving group identification, but also realizes the adaptive allocation of control rights, effectively reduces the control rights conflict in the process of human-machine co-driving, while ensuring driving safety and comfort, and providing personalized human-machine co-driving experience for different driving groups.

[0129] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, 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 application.

Claims

1. A method for allocating human-machine co-driving control rights considering the characteristics of a driving group, characterized in that: include: The vehicle operation data is collected through the vehicle-mounted sensors, and the following vehicle distance data, lateral displacement data, vehicle acceleration data and driving operation frequency data are extracted from the vehicle operation data to obtain driving behavior characteristic data; A deep neural network model is established using driving behavior feature data, and a driving group recognition model is constructed based on a dual-path architecture of manually extracted features and CNN convolutional features; Classify and identify drivers' real-time behaviors based on the driving group recognition model, and generate general driving group labels, aggressive driving group labels, and novice driving group labels; The corresponding control parameters are obtained according to the driving group labels, and the vehicle lateral comprehensive deviation is calculated based on the normalized weighted combination of the preview point lateral deviation and the yaw angle deviation; Using the vehicle's lateral comprehensive deviation and the driver's steering torque, a mathematical model for control right allocation is established through a double exponential function to obtain the intelligent system control weight coefficient and the driver control weight coefficient, including: Establishing a first exponential function according to the vehicle's lateral comprehensive deviation to generate a deviation weight factor; The second exponential function is established using the driver's steering torque to obtain the steering weight factor; The deviation weight factor and the steering weight factor are combined as a function to obtain a control weight reference value; Calculate the initial weight of the intelligent system according to the control weight reference value to obtain the control weight coefficient of the intelligent system; Calculate the initial weight of the driver based on the complementary relationship between the intelligent system and the driver's weight; Normalize the initial weight of the driver to obtain the driver control weight coefficient; According to the driving group label and the group identification frequency data within the time window, the intelligent system control weight coefficient and the driver control weight coefficient are dynamically weighted and combined to generate the control right allocation result.

2. The method for allocating human-machine co-driving control rights considering driving group characteristics according to claim 1 is characterized in that: The vehicle operation data is collected by the vehicle-mounted sensor, and the following vehicle distance data, lateral displacement data, vehicle acceleration data and driving operation frequency data are extracted from the vehicle operation data to obtain driving behavior characteristic data, including: Acquire raw driving data of the vehicle during driving by using a millimeter-wave radar and a visual sensor, and extract basic driving parameters from the raw driving data; Perform multi-sensor data fusion processing on the basic driving parameters to generate real-time following vehicle distance data, lateral displacement data and vehicle acceleration data; Based on the real-time following vehicle distance data and lateral displacement data, the number of emergency braking, the number of emergency acceleration and the number of emergency turning are calculated by a sliding time window algorithm to obtain driving operation frequency data; Using the vehicle acceleration data and the driving operation frequency data, a data normalization algorithm is used to perform feature normalization processing to obtain standardized driving features; Performing data preprocessing on the standardized driving features via a CAN bus to generate a preprocessing feature sequence; The preprocessed feature sequence is subjected to feature combination mapping operation to output the driving behavior feature data.

3. The method for allocating human-machine co-driving control rights considering driving group characteristics according to claim 1 is characterized in that: The method of establishing a deep neural network model using the driving behavior feature data and constructing a driving group recognition model based on a dual-path architecture of manually extracted features and CNN convolutional features includes: Inputting the driving behavior feature data into a pre-trained artificial feature extraction layer to generate an artificial feature vector; Extracting features from the driving behavior feature data through a CNN convolutional layer to obtain a CNN feature vector; The artificial feature vector and the CNN feature vector are combined in series to obtain a fused feature vector; Performing deep neural network training based on the fused feature vector to obtain deep feature representation; Using the deep feature representation and prior knowledge to perform knowledge transfer learning to form knowledge enhancement features; The knowledge enhancement features are classified and learned through a multi-layer perceptron to construct the driving group recognition model.

4. The method for allocating human-machine co-driving control rights considering driving group characteristics according to claim 1 is characterized in that: The method of classifying and judging the real-time behaviors of drivers according to the driving group recognition model to generate general driving group labels, aggressive driving group labels and novice driving group labels includes: Extracting features from real-time driving behavior data using the driving group recognition model to generate a real-time feature vector; Calculate the score probability distribution according to the real-time feature vector to form a group category probability value; Using the group category probability value to set a threshold for multi-classification discrimination, and output an initial group label; Performing time series smoothing processing on the initial group labels to obtain smoothed group labels; Perform group category mapping based on the smoothed group labels to obtain standardized group labels; The standardized group labels are subjected to label fusion processing to generate the general driving group label, the aggressive driving group label and the novice driving group label.

5. The method for allocating human-machine co-driving control rights considering driving group characteristics according to claim 1 is characterized in that: The step of acquiring corresponding control parameters according to the driving group label and calculating the vehicle lateral comprehensive deviation based on the normalized weighted combination of the preview point lateral deviation and the yaw angle deviation includes: Extract the preview point position data from the vehicle status information to obtain the real-time coordinates of the preview point; Calculate the original value of the lateral deviation based on the real-time coordinates of the preview point to generate the lateral deviation of the preview point; The original value of the yaw angle is calculated using the vehicle posture data to obtain the yaw angle deviation; Normalizing the preview point lateral deviation and the yaw angle deviation to obtain a normalized deviation value; Selecting a corresponding weighting coefficient according to the driving group label to generate a deviation weighting parameter; The normalized deviation value and the deviation weighting parameter are combined and calculated to obtain the vehicle lateral comprehensive deviation.

6. The method for allocating human-machine co-driving control rights considering driving group characteristics according to claim 1 is characterized in that: The step of dynamically weighting and combining the intelligent system control weight coefficient and the driver control weight coefficient according to the driving group label and the group identification frequency data in the time window to generate a control right allocation result includes: Count the group recognition results based on the time sliding window to generate group recognition frequency data; Calculating a dynamic weight coefficient using the group identification frequency data to obtain a group weight parameter; Performing a weighted combination of the intelligent system control weight coefficient and the group weight parameter to obtain a final weight of the intelligent system; Performing a weighted combination on the driver control weight coefficient and the group weight parameter to generate a final driver weight; Calculating the control right distribution according to the final weight of the intelligent system and the final weight of the driver; The control right allocation calculation result is normalized to obtain the control right allocation result.

7. A device for allocating human-machine co-driving control rights considering the characteristics of a driving group, characterized in that: The human-machine co-driving control rights allocation device considering the driving group characteristics includes: An extraction module is used to collect vehicle operation data through vehicle-mounted sensors, extract vehicle-to-vehicle distance data, lateral displacement data, vehicle acceleration data, and driving operation frequency data from the vehicle operation data, and obtain driving behavior characteristic data; A construction module is used to establish a deep neural network model using the driving behavior feature data, and to construct a driving group recognition model based on a dual-path architecture of manually extracted features and CNN convolutional features; A discrimination module, used to classify and discriminate the real-time behaviors of drivers according to the driving group recognition model, and generate general driving group labels, aggressive driving group labels and novice driving group labels; A combination module, used for obtaining corresponding control parameters according to the driving group label, and calculating the vehicle lateral comprehensive deviation based on the normalized weighted combination of the preview point lateral deviation and the yaw angle deviation; A distribution module, used to establish a control right distribution mathematical model by using the vehicle lateral comprehensive deviation and the driver's steering torque through a double exponential function to obtain an intelligent system control weight coefficient and a driver control weight coefficient; A weighting module, for dynamically weighting and combining the intelligent system control weight coefficient and the driver control weight coefficient according to the driving group label and the group identification frequency data within the time window to generate a control right allocation result; The allocation module is used to establish a first exponential function according to the vehicle's lateral comprehensive deviation to generate a deviation weight factor; use the driver's steering torque to establish a second exponential function to obtain a steering weight factor; perform functional combination of the deviation weight factor and the steering weight factor to obtain a control weight reference value; calculate the initial weight of the intelligent system according to the control weight reference value to obtain the intelligent system control weight coefficient; calculate the driver's initial weight based on the complementary relationship between the intelligent system and the driver's weight; and normalize the driver's initial weight to obtain the driver's control weight coefficient.

Images