Intelligent driving method suitable for multiple road conditions

Abstract

The invention discloses an intelligent driving method suitable for multiple road conditions, and belongs to the field of intelligent driving, and the method comprises the following steps: S1, multi-modal coupling sensing data collection and fusion processing; s2, performing driver-road condition-risk dynamic coupling evaluation; s3, performing weather-road condition adaptive trajectory planning; s4, driver-own vehicle-road condition cooperative braking and control execution; and S5, closed-loop feedback and parameter iterative updating are carried out. The intelligent driving method suitable for multiple road conditions can adapt to complex road conditions (such as low-visibility and low-friction-coefficient road surfaces), considers driving safety, stability and energy efficiency, and has a wide application prospect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent driving, and in particular to an intelligent driving method suitable for multiple road conditions. BACKGROUND

[0002] With the transformation of the global automobile industry towards electrification, intelligentization and networking, L2+ intelligent driving systems have become a standard for mass-produced vehicles and are rapidly evolving towards L3 conditional automatic driving. With the development of intelligent driving technology, intelligent driving systems need to break through core bottlenecks such as insufficient environmental perception robustness, weak coordination between driver state and vehicle control, and poor adaptability of multi-scenario braking strategies, in order to cope with the complex challenges of multiple road conditions such as urban roads, highways and rural roads.

[0003] Current main intelligent driving solutions use CNN (Convolutional Neural Network) or traditional SVM (Support Vector Machine) to identify driver states such as alertness, distraction and drowsiness based on fNIRS brain oxygen signals and EEG brain electrical signals. There are defects such as dependence on large-scale labeled sample training, weak generalization ability for small samples and cross-scenarios, and failure to adjust the recognition strategy dynamically in combination with driving scenarios. SUMMARY

[0004] The purpose of the present application is to provide a solution to the above technical problems.

[0005] To achieve the above purpose, the present application provides an intelligent driving method suitable for multiple road conditions, comprising the following steps: S1, multi-modal coupled perception data acquisition and fusion processing: synchronously acquiring environment, driver state and vehicle state data, suppressing noise through 4th order Butterworth low-pass filter, extracting multi-modal features, and then using joint mutual information weighted fusion mechanism to dynamically allocate each modal feature weight, and outputting fusion perception data containing environment features, driver state and vehicle state; S2, driver-road condition-risk dynamic coupling evaluation: based on the fusion perception data output by S1, the driver state is identified and the distraction risk is quantified through an improved SVM-JKNN hybrid model, the dynamic visibility and road friction coefficient are obtained by combining the Koschmieder scattering model and Monte Carlo simulation, a multi-dimensional coupled risk model is constructed using Bayesian network, the total risk value and key risk factors are quantified by introducing low-visibility risk field LVRFF, and the risk evaluation result and risk level are output; S3, weather-road condition adaptive trajectory planning: based on the risk evaluation result output by S2 and the fusion perception data, a five-order polynomial lateral trajectory and a cubic acceleration longitudinal trajectory coupling framework is used for different road conditions, an optimal trajectory is generated by combining model predictive control MPC, and the optimal trajectory is generated based on the optimal trajectory and the road condition; S4, Driver-Vehicle-Road Condition Cooperative Braking and Control Execution: Based on the optimal trajectory scheme and braking strategy output by S3, the MPC-Dual PID composite control algorithm is used to track the trajectory, and when the total risk value exceeds the set risk warning threshold, seat vibration, voice warning and emergency intervention are triggered, and steering, acceleration and braking control signals are output. S5. Closed-loop feedback and parameter iterative update: Deploy IoT sensors to collect data on vehicle usage status, road condition changes, and driver behavior. Use gradient descent algorithm to optimize the perception fusion weights of S1 and the Bayesian network parameters of S2 to achieve closed-loop optimization.

[0006] Therefore, the present invention employs the above-mentioned intelligent driving method applicable to multiple road conditions, and has the following beneficial effects: 1. By adopting the approach of "data type preprocessing (such as physiological data filtering and image defogging) + improved YOLOv10 multi-scale flexible convolution + JMI dynamic weight fusion", the problem of "information fragmentation and insufficient noise suppression" of traditional single sensor / fixed weight fusion is solved. In complex scenarios such as low visibility and low friction road surface, the perception accuracy is improved by more than 30% compared with the traditional solution. 2. By using Bayesian networks to couple and assess "driver's physiological state (distraction / fatigue), environmental risk (visibility / road friction), and collision risk (TTC)," rather than the traditional "single risk factor superposition," the quantitative results are more in line with the actual scenario (e.g., the risk superposition effect of low visibility + driver distraction is more accurate), and the risk misjudgment rate is reduced by 25%. 3. Dynamically allocate the hydraulic / regenerative braking ratio according to the road surface type (icy / mixed / dry), ensuring safety on low-friction roads while improving energy recovery efficiency by 15%-20% on dry roads; 4. It adopts a composite control architecture of "MPC pre-control + dual PID precision control" and is equipped with a three-level warning system of "voice-vibration-forced braking". This ensures the accuracy of trajectory tracking (lateral position deviation ≤0.2m) and can quickly intervene in high-risk scenarios, solving the defects of traditional single PID control such as "slow response and integral saturation when large deviation".

[0007] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0008] Figure 1 This is a flowchart of an intelligent driving method applicable to multiple road conditions as described in this invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0010] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0011] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0012] like Figure 1 As shown, an intelligent driving method applicable to various road conditions includes the following steps: S1. Multimodal Coupled Sensing Data Acquisition and Fusion Processing: Simultaneously acquire environmental, driver status, and vehicle status data. Suppress noise through 4th-order Butterworth low-pass filtering. After extracting multimodal features, adopt a joint mutual information weighted fusion mechanism to dynamically allocate the weights of each modal feature and output fused sensing data containing environmental features, driver status, and vehicle status. S2, Driver-Road Condition-Risk Dynamic Coupling Assessment: Based on the fused perception data output from S1, the driver's state is identified and distraction risk is quantified by improving the SVM-JKNN hybrid model. Dynamic visibility and road friction coefficient are obtained by combining the Koschmieder scattering model and Monte Carlo simulation. A multi-dimensional coupled risk model is constructed using a Bayesian network. The low visibility risk field force LVRFF is introduced to quantify the total risk value and key risk factors, and the risk assessment results and risk level are output. S3, Weather-Road Condition Adaptive Trajectory Planning: Based on the risk assessment results output by S2 and the fused perception data, a fifth-order polynomial lateral trajectory and a cubic acceleration longitudinal trajectory coupling framework are adopted for different road conditions. The optimal trajectory is generated by combining model prediction control MPC, and road conditions are considered based on the optimal trajectory. S4, Driver-Vehicle-Road Condition Cooperative Braking and Control Execution: Based on the optimal trajectory scheme and braking strategy output by S3, the MPC-Dual PID composite control algorithm is used to track the trajectory, and when the total risk value exceeds the set risk warning threshold, seat vibration, voice warning and emergency intervention are triggered, and steering, acceleration and braking control signals are output. S5. Closed-loop feedback and parameter iterative update: Deploy IoT sensors to collect data on vehicle usage status, road condition changes, and driver behavior. Use gradient descent algorithm to optimize the perception fusion weights of S1 and the Bayesian network parameters of S2 to achieve closed-loop optimization.

[0013] The environmental data mentioned in step S1 includes roadside CCTV images. Millimeter-wave radar signals LiDAR point cloud And ambient light sensor data ; Driver status data includes fNIRS brain oxygenation signals and EEG brain signals ; Vehicle status data Including vehicle speed Acceleration, braking pressure Remaining battery power and heading angle .

[0014] Step S1 specifically includes the following steps: S11. By using timestamp alignment technology, environmental, driver, and vehicle status data are collected synchronously to obtain time-aligned multi-source raw data: ; S12. Preprocessing and noise suppression: Targeting fNIRS brain oxygenation signals and EEG brain signals A fourth-order Butterworth low-pass filter is used to suppress physiological noise and environmental interference; Rain and fog noise is removed from roadside CCTV images using a dark channel prior algorithm. The vehicle status data is smoothed using a recursive least squares method. Outlier points are removed from lidar point cloud data using statistical filtering. The denoised preprocessed data is obtained: ; These represent the denoised roadside CCTV image, millimeter-wave radar signal, lidar point cloud, fNIRS brain oxygenation signal, EEG brain signal, vehicle status data, and ambient light sensor data, respectively. In this embodiment, the following are used: roadside CCTV images (30fps, 5760×1080 resolution), millimeter-wave radar (77GHz, maximum detection range 300m, sampling frequency 100Hz), lidar point cloud (sampling frequency 10Hz), 24-channel fNIRS device (sampling frequency 50Hz, emission wavelength 760nm / 840nm, interpole spacing 35mm, covering the prefrontal cortex), EEG brainwave device (sampling frequency 250Hz, collecting δ / θ / α / β / γ waves), OBD-II interface (sampling frequency 10Hz, collecting engine speed, vehicle speed v, acceleration a, braking pressure Pb), and ambient light sensor (sampling frequency 5Hz).

[0015] S13, Multimodal Feature Extraction; Extracting environmental features from preprocessed environmental data: Extracting obstacle locations based on an improved YOLOv10 multi-scale flexible convolutional module. ,size and speed Dynamic visibility was extracted using the Koschmieder scattering model combined with lane line calibration. Simultaneously, the road surface friction coefficient is extracted based on Monte Carlo simulation. ; Extracting driver features from preprocessed driver state data: from preprocessed fNIRS brain oxygenation signals Extract the mean oxyhemoglobin feature from the 30s non-overlapping window. Characteristics of mean deoxyhemoglobin Temporal features were extracted using the sliding window method; simultaneously, the Welch method was used to extract features from the preprocessed EEG signals. Extracting coverage Wave, Wave, Wave, wave and Power spectral density characteristics of wave energy proportion ; From preprocessed vehicle status data Extracting vehicle features: vehicle speed Longitudinal acceleration lateral acceleration Braking pressure Remaining battery power Heading angle ; Obtain multimodal raw features , These represent the original environmental feature set, the original driver feature set, and the original vehicle state feature set, respectively. S14. Multimodal Feature Fusion: Based on joint mutual information, calculate the correlation between each feature and the target output, combined with road condition type. By dynamically adjusting the feature adaptation weights, adaptive weighted fusion of multimodal features is achieved, resulting in fused perceptual data. .

[0016] The improved YOLOv10 multi-scale flexible convolutional module described in step S13 includes an input adaptation processing layer, a multi-scale parallel convolutional branch layer, a flexible weight learning layer, a road condition adaptive weight adjustment layer, a feature fusion layer, and an output adaptation processing layer arranged sequentially. The input adaptation processing layer receives the preprocessed roadside CCTV image data output from step S12, and performs normalization and contrast enhancement on it sequentially to obtain a denoised and normalized feature map. The feature map is processed by a 1×1 convolution kernel. The number of channels has been uniformly adjusted to 64, and the output input features have been standardized. ; The multi-scale parallel convolutional branch layer (which uses three independent parallel convolutional branches to extract target features at different scales simultaneously) includes a 1×1 convolutional branch, a 3×3 convolutional branch, and a 5×5 convolutional branch. All three branches use the ReLU activation function to enhance non-linear expressive power, and the output feature maps of each branch have the same dimension. Flexible weight learning layer, used to introduce learnable basic weight parameters. The initial value of the basic weight parameter is set to 1 / 3. During the model training phase, it is optimized using the stochastic gradient descent algorithm based on a multi-road dataset. The optimization objective is to minimize the total detection loss, which combines classification loss, regression loss, and confidence loss. ; in, ; ; ; In the formula, This represents the total loss value; , and All represent the weighting coefficients for the loss term; , and These represent classification loss, regression loss, and confidence loss, respectively. This represents the total number of positive samples; This represents the number of grids in the output feature map of the improved YOLOv10 multi-scale flexible convolutional module. This indicates the number of anchor frames preset for each grid. This represents the positive sample indicator function, i.e., if the first... The first grid If an anchor box matches the real target, the value is 1; otherwise, it is 0. This indicates the weight of road condition adaptation, and , Indicates the road surface friction coefficient and the fit coefficient. This represents the critical road surface friction coefficient. This represents the visibility adaptation coefficient. Indicates the safe stopping sight distance, and , This represents the driver's reaction time coefficient. The fitting term representing the road friction coefficient indicates the braking distance. The slope influence coefficient represents the braking distance. Indicates the road surface slope; Indicates the total number of target categories; Indicates the first The first grid Each anchor box corresponds to the true category of the target; The model predicts the first The first grid Each anchor frame corresponds to a target belonging to a category. The probability of; This represents the intersection-union ratio (IoU) between the predicted bounding box and the ground truth bounding box. Indicates the center of the prediction box Center of the real frame The square of the Euclidean distance; This represents the square of the diagonal length of the smallest rectangle containing both the predicted and ground truth boxes. This indicates the weight used to balance the aspect ratio loss; This indicates the difference in aspect ratio between the predicted bounding box and the ground truth bounding box; This represents the total number of anchor frames across all grids, and ; Indicates the first The first grid The probability that each anchor box contains the target; Indicates the weight of negative samples; This represents the negative sample indicator function, which is 1 if the anchor box does not match the true target, and 0 otherwise. And the weights are constrained by the Softmax function to satisfy... , These represent the base weights of the 1×1 convolution branch, the 3×3 convolution branch, and the 5×5 convolution branch, respectively. This represents the index of the convolution branch, enabling adaptive learning of the feature contribution of each branch; The adaptive weighting layer for road conditions is used to adjust the base weight parameters based on real-time road condition parameters. Dynamic adjustments are made to determine the final weights that best suit the current road conditions. ; The formula for dynamically adjusting the basic weights is as follows: ; In the formula, Indicates the first The final weights after normalization of each convolutional branch. ; The feature fusion layer is used to fuse the output feature maps of the three convolutional branches using an element-wise weighted summation method, and the fusion expression is as follows: ; In the formula, This represents the fused feature map; , and These represent the final weights after normalization for the 1×1 convolutional branch, the 3×3 convolutional branch, and the 5×5 convolutional branch, respectively. , and These represent the feature maps output by the 1×1 convolutional branch, the 3×3 convolutional branch, and the 5×5 convolutional branch, respectively. The output adaptation layer is used to process the fused feature maps. The signal undergoes a nonlinear transformation using an activation function, and the output is fed into the C2f module of the YOLOv10 backbone network for subsequent downsampling and semantic feature enhancement.

[0017] Step S14 specifically includes the following steps: S141. The multimodal raw features output in step S13 The features are categorized and organized, and the target output corresponding to each category is clearly defined. A feature-target mapping relationship is then established, i.e., environmental features. The corresponding target output is the environmental risk level. Driver characteristics The corresponding target output is the driver's level of distraction or drowsiness. Features of the vehicle The target output is the vehicle's safe operating status. ; Obtain the classification feature set and target output set ;in, ; S142. Calculate the JMI value between each classified feature and its corresponding target output: ; In the formula, Indicates the first Class features With the corresponding target output The joint mutual information value; Representation of features Information entropy; Indicates the target output Information entropy; Representation of features With the corresponding target output The joint entropy; Obtain the JMI value set for each feature. ; S143, According to road condition type The preset road condition-feature adaptation rule table is used to assign adaptation weights to various features, resulting in a road condition adaptation weight set for each feature. , Represented as the first Adaptation weights assigned to class features; S144. Calculate the initial weights: ; In the formula, Indicates the first Initial weights for class features; Simultaneously, by normalizing the fusion weights, the sum of the fusion weights of all features is ensured to be 1, resulting in a normalized fusion weight set. ; S145. Multiply each classification feature by its corresponding normalized fusion weight and sum them to obtain the fusion feature: ; In the formula, Indicates the first Class fusion features, These represent the indices for environmental fusion features, driver fusion features, and vehicle fusion features, respectively. Indicates the first The subset of original features corresponding to the class fusion features; S146, Integration of Environmental Characteristics Driver fusion characteristics and the characteristics of vehicle integration This forms the final fused sensing data. .

[0018] Step S2 specifically includes the following steps: S21. Mean oxyhemoglobin characteristics in the fused sensing data based on the output of S1 Characteristics of mean deoxyhemoglobin and power spectral density characteristics Input the JKNN-SVM model, output the driver state classification results, and then calculate the risk value by combining the state duration; S211, Mean characteristics of paraoxyhemoglobin Characteristics of mean deoxyhemoglobin and power spectral density characteristics Z-score standardization is performed to obtain the standardized feature vector. ; S212. The similarity between the test features and all samples in the training set is calculated using Jaccard distance, and the top-ranked features are then selected. One nearest neighbor sample; S213, the output of S212 The nearest neighbor sample set optimizes the training sample distribution of the SVM, and the RBF kernel function is used to construct the classifier. The penalty coefficient and kernel function parameters of the SVM are determined through Bayesian parameter optimization, and the driver state level is output. : ; in, ; In the formula, Indicates the SVM decision score; Represents the standardized feature vector to be classified; Represent the Lagrange multipliers of SVM; The labels representing the training samples are the driver's state levels. Represents the kernel function, and , This represents the width parameter of the RBF kernel function. This represents the feature vector of a sample in the training set; Indicates the bias term; Indicates the preceding Total number of nearest neighbor samples; S214. Driver status level obtained based on classification and state duration Quantifying driver distraction risk using the Sigmoid function : ; In the formula, Indicates the risk growth coefficient; Indicates the threshold for the duration of the state; S22. Quantify environmental visibility risks separately. Road surface friction risk and collision risk : ; ; ; In the formula, This indicates the time of collision between the vehicle and the obstacle in front, and , Indicates the relative distance between the vehicle and the obstacle; Indicates the collision warning threshold; S23. Construct a coupled risk model based on Bayesian networks to address driver distraction risk. Environmental visibility risk Road surface friction risk Collision risk As an input node, total risk As the output node, the coupling relationship between various risk factors is quantified using a conditional probability table to obtain the total risk. and risk level ; in, ; In the formula, Indicates that given network parameters Under these conditions, driver distraction risk Visibility risk Road surface friction risk Collision risk The probability of them happening simultaneously; Indicates the parameters of the Bayesian network; Indicates; in the Parent node of each risk factor Status and network parameters Next, the The probability of each risk factor occurring These respectively represent the risk of driver distraction. Environmental visibility risk Road surface friction risk Collision risk index.

[0019] Step S3 specifically includes the following steps: S31. Normalize the road surface friction coefficient Available lane distance Speed ​​deviation Risk level Road curvature As indicators, weights are dynamically assigned to each indicator according to a pre-defined scenario-weight mapping table. The normalized indicators and the assigned dynamic weights are then substituted into the comprehensive evaluation function to calculate the sum score for each candidate lane. : ; In the formula, , , , and These represent the normalized road surface friction coefficient, lane availability distance, vehicle speed deviation, risk level, and road curvature, respectively. , , , , These are respectively represented as the normalized road surface friction coefficient, lane availability distance, vehicle speed deviation, risk level, and dynamic weights of road curvature distribution; S32, Sum score of all candidate lanes Sort the lanes in descending order, exclude lanes that are not allowed to proceed, and select the lane with the highest score as the optimal decision. ; S33. Extract shallow spatial features and deep semantic features using the MAFPN fusion step S13: ; In the formula, This represents the final fused feature output by the improved YOLOv10 multi-scale flexible convolutional module; S34. Predicting the future using extended Kalman filtering. obstacle movement trajectory within seconds This provides dynamic obstacle constraints for trajectory planning; among them, express The location of obstacles at all times. express The speed of the obstacle at any given moment; S35. Use fifth-order polynomial programming for the lateral trajectory: ; In the formula, express The vehicle's lateral position at all times; , , , , and All represent the coefficients of a fifth-order polynomial; Represents a time variable; By utilizing boundary constraints and introducing lateral acceleration constraints under multiple road conditions, the coefficients of the fifth-order polynomial are solved: Boundary constraints: ; Lateral acceleration constraints under multiple road conditions: ; In the formula, Indicates the initial lateral position of the vehicle; Indicates the current lateral position of the vehicle; Indicates the horizontal position of the target; Indicates the target lateral position of the vehicle at the moment the trajectory ends; This represents the lateral velocity of the vehicle at the initial moment; This represents the lateral velocity of the vehicle at the moment the trajectory ends; This represents the lateral acceleration of the vehicle at the initial moment; This represents the lateral acceleration of the vehicle at the moment the trajectory ends; express Horizontal velocity at all times; Indicates the maximum lateral acceleration; Represents gravitational acceleration; get , , , , , ; S36. Use a cubic acceleration model to plan the longitudinal trajectory, combined with the risk level. and the trajectory of the obstacle The longitudinal acceleration is dynamically adjusted to match the longitudinal velocity with the lateral trajectory. The expression for the cubic acceleration model is as follows: ; In the formula, Indicates the future The longitudinal speed of the vehicle at all times; This indicates the current longitudinal speed of the vehicle; Indicates the initial longitudinal acceleration; Indicates the rate of change of longitudinal acceleration; Constraints: Speed ​​constraints: ; Acceleration constraints: ; Collision avoidance constraints: ; In the formula, Indicates the minimum safe speed; express longitudinal acceleration at all times ; Indicates the maximum longitudinal acceleration; express The longitudinal position of the vehicle at all times; Indicates the length of the vehicle; express The longitudinal position of the obstacle ahead at any given moment; Indicates a safe distance; S37. Construct a longitudinally and laterally coupled frame, and adjust the longitudinal velocity and lateral acceleration through coupling constraints: ; ; Coupling constraints: ; In the formula, This indicates the longitudinal velocity after coupling adjustment; This represents the lateral acceleration after coupling adjustment; This indicates the lateral acceleration before adjustment. ; S38. Model predictive control is used to smooth and optimize the coupled longitudinal and lateral trajectories, with trajectory smoothness, safety, and efficiency as optimization objectives. The optimization weights are dynamically adjusted to adapt to multiple road conditions to obtain the initial trajectory of the vehicle. : ; Among them, the model predictive control optimization objective function The expression is as follows: ; Constraints: Vertical constraints: and ; Lateral constraints: and ; Safety constraints: ; In the formula, , , , Indicates the optimization weights; Represents the optimal decision The optimal speed in the system; Indicates the minimum distance between the vehicle and the obstacle; These represent the lateral positions of the left and right boundaries of the lane, respectively. S39. With trajectory smoothness, safety, and energy efficiency as optimization objectives, the initial trajectory is iteratively optimized using the MPC algorithm. Output the optimal trajectory , Indicates the optimal position. Indicates the optimal speed. This represents the optimal longitudinal acceleration. This represents the optimal lateral acceleration. Indicates the optimal heading angle; The multi-objective optimization function is expressed as follows: ; Constraints: ; In the formula, , and Both represent optimized weights; Indicates the real-time distance between the vehicle and obstacles in front; S310, From the optimal trajectory Extract the target deceleration at the current moment. Deceleration duration and speed difference : ; ; ; In the formula, Indicates the current vehicle speed; express The target vehicle speed after time step is determined by the optimal trajectory. Obtain; Indicates the default deceleration time domain; This indicates the maximum permissible deceleration, and , Indicates the safety redundancy coefficient; S311, Based on road surface friction coefficient Total Risk Value The optimal regenerative braking efficiency is calculated by adjusting for the vehicle's remaining battery power. : ; in, ; In the formula, Indicates the basic regeneration efficiency; Indicates the preset regenerative braking target efficiency; , and All represent weighting coefficients; S312, Based on total braking intensity and optimal regenerative braking efficiency Distribute hydraulic braking intensity and regenerative braking strength : ; ; in, ; Road condition adaptation constraints: when hour, and ; when hour, and ; when hour, and ; In the formula, Represents the optimal trajectory The target deceleration; This represents the corrected deceleration time domain; S313, targeting On icy and snowy roads, a closed-loop slip ratio control system is introduced to control wheel slip ratio. The optimal range is 10%-15%, which can be achieved through adjustments. and The dynamic response speed prevents wheel lock-up or skidding. ; ; in, ; ; In the formula, and These represent the hydraulic braking strength and regenerative braking strength after slip ratio correction, respectively; This represents the slip ratio correction factor; Indicates real-time wheel speed; Indicates the rolling radius of the wheel; S314, in conjunction with risk level Adjust the response priority of the braking strategy; when When it is high risk, , ; when When it is medium risk, , ; when When it is low risk, , ; In the formula, and These represent the final hydraulic braking strength and regenerative braking strength after risk adaptation, respectively. S315, Output Braking Strategy , This represents the preset target slip ratio. In this embodiment, when... hour, Take 5%-8%; when hour, Take 9%-12%; hour, Take 13%-15%.

[0020] Step S4 specifically includes the following steps: S41, the optimal trajectory output by S3 Risk assessment results of S2 , The dataset is obtained by performing timestamp alignment and noise suppression on the sensing data of S1. ; in, These represent the optimal trajectory, total risk, driver status level, road friction coefficient, and vehicle status after synchronous preprocessing, respectively. S42. Aiming at the optimal trajectory after synchronization, based on the vehicle dynamics model and future... The deviation of each prediction step is used to construct the MPC objective function, and the pre-heading angle is generated under road condition constraints. and pre-throttle opening ; Wherein, the MPC objective function The expression is as follows: ; Control constraints: ,and ; Dynamic constraints: ; In the formula, , , , , , All represent deviation weights; , , , and These represent the prediction steps. The longitudinal position deviation, lateral position deviation, longitudinal velocity deviation, longitudinal acceleration deviation, and heading angle deviation of the vehicle at any given time; Indicates control over the time domain; Indicates the prediction time domain; and Both represent the smoothing weight of the control quantity; express Increment of heading angle at any moment; express The increment of throttle opening at any given moment; and These represent the maximum and minimum heading angles, respectively. express Heading angle at any moment; express Throttle opening at all times; express lateral acceleration at all times; S43. Utilize dual PID controllers to track lateral position and longitudinal velocity, where the lateral PID controller uses... and Input: , Output: Final heading angle Vertical PID and For output, output the final valve opening. ; The PID control expression for the lateral position is as follows: ; in, ; In the formula, , and These represent the proportional, integral, and derivative parameters of the lateral position PID control, respectively. express lateral positional deviation at any moment; and They represent Time and The horizontal integral term at time step; express Rate of change of position deviation at time; Indicates the integral separation threshold; Indicates the sampling time interval; The longitudinal velocity PID control expression is as follows: ; in, ; In the formula, , and These represent the proportional, integral, and derivative parameters of the longitudinal velocity PID control, respectively. express Momentary velocity deviation; and They represent Time and The longitudinal integral term at time step; express Rate of change of velocity deviation at any given time; and These represent the upper and lower limits of saturation for vertical integration, respectively. Represents; represents a saturation function; S44. After integrating and limiting the braking signal and trajectory tracking signal, output the final steering, acceleration and braking signals; Among them, steering control: Acceleration control: ; Braking control: ; And during the driving process, based on the total risk after pretreatment and driver status level Output the following early warning strategy: when or When it enters the mild warning mode, it will output a voice warning. when or When the system enters a severe warning mode, it will output seat vibration and voice warning. when When entering emergency warning mode, it will be forcibly activated. Upgraded to The longitudinal speed decreased to ; express The optimal longitudinal speed of the vehicle at any given time is taken from the optimal speed. ; Furthermore, when entering emergency mode, braking control is set to the highest priority.

[0021] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent driving method applicable to multiple road conditions, characterized in that: Includes the following steps: S1. Multimodal Coupled Sensing Data Acquisition and Fusion Processing: Simultaneously acquire environmental, driver status, and vehicle status data. Suppress noise through 4th-order Butterworth low-pass filtering. After extracting multimodal features, adopt a joint mutual information weighted fusion mechanism to dynamically allocate the weights of each modal feature and output fused sensing data containing environmental features, driver status, and vehicle status. S2, Driver-Road Condition-Risk Dynamic Coupling Assessment: Based on the fused perception data output from S1, the driver's state is identified and distraction risk is quantified by improving the SVM-JKNN hybrid model. Dynamic visibility and road friction coefficient are obtained by combining the Koschmieder scattering model and Monte Carlo simulation. A multi-dimensional coupled risk model is constructed using a Bayesian network. The low visibility risk field force LVRFF is introduced to quantify the total risk value and key risk factors, and the risk assessment results and risk level are output. S3, Weather-Road Condition Adaptive Trajectory Planning: Based on the risk assessment results output by S2 and the fused perception data, a fifth-order polynomial lateral trajectory and a cubic acceleration longitudinal trajectory coupling framework are adopted for different road conditions. The optimal trajectory is generated by combining model prediction control MPC, and road conditions are considered based on the optimal trajectory. S4, Driver-Vehicle-Road Condition Cooperative Braking and Control Execution: Based on the optimal trajectory scheme and braking strategy output by S3, the MPC-Dual PID composite control algorithm is used to track the trajectory, and when the total risk value exceeds the set risk warning threshold, seat vibration, voice warning and emergency intervention are triggered, and steering, acceleration and braking control signals are output. S5. Closed-loop feedback and parameter iterative update: Deploy IoT sensors to collect data on vehicle usage status, road condition changes, and driver behavior. Use gradient descent algorithm to optimize the perception fusion weights of S1 and the Bayesian network parameters of S2 to achieve closed-loop optimization.

2. The intelligent driving method applicable to multiple road conditions according to claim 1, characterized in that: The environmental data mentioned in step S1 includes roadside CCTV images. Millimeter-wave radar signals LiDAR point cloud And ambient light sensor data ; Driver status data includes fNIRS brain oxygenation signals and EEG brain signals ; Vehicle status data Including vehicle speed Acceleration, braking pressure Remaining battery power and heading angle .

3. The intelligent driving method applicable to multiple road conditions according to claim 2, characterized in that: Step S1 specifically includes the following steps: S11. By using timestamp alignment technology, environmental, driver, and vehicle status data are collected synchronously to obtain time-aligned multi-source raw data: ; S12. Preprocessing and noise suppression: Targeting fNIRS brain oxygenation signals and EEG brain signals A fourth-order Butterworth low-pass filter is used to suppress physiological noise and environmental interference; Rain and fog noise is removed from roadside CCTV images using a dark channel prior algorithm. The vehicle status data is smoothed using a recursive least squares method. Outlier points are removed from lidar point cloud data using statistical filtering. The denoised preprocessed data is obtained: ; These represent the denoised roadside CCTV image, millimeter-wave radar signal, lidar point cloud, fNIRS brain oxygenation signal, EEG brain signal, vehicle status data, and ambient light sensor data, respectively. S13, Multimodal Feature Extraction; Extracting environmental features from preprocessed environmental data: Extracting obstacle locations based on an improved YOLOv10 multi-scale flexible convolutional module. ,size and speed Dynamic visibility was extracted using the Koschmieder scattering model combined with lane line calibration. Simultaneously, the road surface friction coefficient is extracted based on Monte Carlo simulation. ; Extracting driver features from preprocessed driver state data: from preprocessed fNIRS brain oxygenation signals Extract the mean oxyhemoglobin feature from the 30s non-overlapping window. Characteristics of mean deoxyhemoglobin Temporal features were extracted using the sliding window method; simultaneously, the Welch method was used to extract features from the preprocessed EEG signals. Extracting coverage Wave, Wave, Wave, wave and Power spectral density characteristics of wave energy proportion ; From preprocessed vehicle status data Extracting vehicle features: vehicle speed Longitudinal acceleration lateral acceleration Braking pressure Remaining battery power Heading angle ; Obtain multimodal raw features , These represent the original environmental feature set, the original driver feature set, and the original vehicle state feature set, respectively. S14. Multimodal Feature Fusion: Based on joint mutual information, calculate the correlation between each feature and the target output, combined with road condition type. By dynamically adjusting the feature adaptation weights, adaptive weighted fusion of multimodal features is achieved, resulting in fused perceptual data. .

4. The intelligent driving method applicable to multiple road conditions according to claim 3, characterized in that: The improved YOLOv10 multi-scale flexible convolutional module described in step S13 includes an input adaptation processing layer, a multi-scale parallel convolutional branch layer, a flexible weight learning layer, a road condition adaptive weight adjustment layer, a feature fusion layer, and an output adaptation processing layer arranged sequentially. The input adaptation processing layer receives the preprocessed roadside CCTV image data output from step S12, and performs normalization and contrast enhancement on it sequentially to obtain a denoised and normalized feature map. The feature map is processed by a 1×1 convolution kernel. The number of channels has been uniformly adjusted to 64, and the output input features have been standardized. ; The multi-scale parallel convolutional branch layer includes a 1×1 convolutional branch, a 3×3 convolutional branch, and a 5×5 convolutional branch. All three branches use the ReLU activation function to enhance non-linear expressive power, and the output feature maps of each branch have the same dimension. Flexible weight learning layer, used to introduce learnable basic weight parameters. The initial value of the basic weight parameter is set to 1 / 3. During the model training phase, it is optimized using the stochastic gradient descent algorithm based on a multi-road dataset. The optimization objective is to minimize the total detection loss, which combines classification loss, regression loss, and confidence loss. ； in, ； ； ； In the formula, This represents the total loss value; , and All represent the weighting coefficients for the loss term; , and These represent classification loss, regression loss, and confidence loss, respectively. This represents the total number of positive samples; This represents the number of grids in the output feature map of the improved YOLOv10 multi-scale flexible convolutional module. This indicates the number of anchor frames preset for each grid. This represents the positive sample indicator function, i.e., if the first... The first grid If an anchor box matches the real target, the value is 1; otherwise, it is 0. This indicates the weight of road condition adaptation, and , Indicates the road surface friction coefficient and the fit coefficient. This represents the critical road surface friction coefficient. This represents the visibility adaptation coefficient. Indicates the safe stopping sight distance, and , This represents the driver's reaction time coefficient. The fitting term representing the road friction coefficient indicates the braking distance. The slope influence coefficient represents the braking distance. Indicates the road surface slope; Indicates the total number of target categories; Indicates the first The first grid Each anchor box corresponds to the true category of the target; The model predicts the first The first grid Each anchor frame corresponds to a target belonging to a category. The probability of; This represents the intersection-union ratio (IoU) between the predicted bounding box and the ground truth bounding box. Indicates the center of the prediction box Center of the real frame The square of the Euclidean distance; This represents the square of the diagonal length of the smallest rectangle containing both the predicted and ground truth boxes. This indicates the weight used to balance the aspect ratio loss; This indicates the difference in aspect ratio between the predicted bounding box and the ground truth bounding box; This represents the total number of anchor frames across all grids, and ; Indicates the first The first grid The probability that each anchor box contains the target; Indicates the weight of negative samples; This represents the negative sample indicator function, which is 1 if the anchor box does not match the true target, and 0 otherwise. And the weights are constrained by the Softmax function to satisfy... , These represent the base weights of the 1×1 convolution branch, the 3×3 convolution branch, and the 5×5 convolution branch, respectively. This represents the index of the convolution branch, enabling adaptive learning of the feature contribution of each branch; The adaptive weighting layer for road conditions is used to adjust the base weight parameters based on real-time road condition parameters. Dynamic adjustments are made to determine the final weights that best suit the current road conditions. ; The formula for dynamically adjusting the basic weights is as follows: ； In the formula, Indicates the first The final weights after normalization of each convolutional branch. ; The feature fusion layer is used to fuse the output feature maps of the three convolutional branches using an element-wise weighted summation method, and the fusion expression is as follows: ； In the formula, This represents the fused feature map; , and These represent the final weights after normalization for the 1×1 convolutional branch, the 3×3 convolutional branch, and the 5×5 convolutional branch, respectively. , and These represent the feature maps output by the 1×1 convolutional branch, the 3×3 convolutional branch, and the 5×5 convolutional branch, respectively. The output adaptation layer is used to process the fused feature maps. The signal undergoes a nonlinear transformation using an activation function, and the output is fed into the C2f module of the YOLOv10 backbone network for subsequent downsampling and semantic feature enhancement.

5. The intelligent driving method applicable to multiple road conditions according to claim 3, characterized in that: Step S14 specifically includes the following steps: S141. The multimodal raw features output in step S13 The features are categorized and organized, and the target output corresponding to each category is clearly defined. A feature-target mapping relationship is then established, i.e., environmental features. The corresponding target output is the environmental risk level. Driver characteristics The corresponding target output is the driver's level of distraction or drowsiness. Features of the vehicle The target output is the vehicle's safe operating status. ; Obtain the classification feature set and target output set ;in, ; S142. Calculate the JMI value between each classified feature and its corresponding target output: ； In the formula, Indicates the first Class features With the corresponding target output The joint mutual information value; Representation of features Information entropy; Indicates the target output Information entropy; Representation of features With the corresponding target output The joint entropy; Obtain the JMI value set for each feature. ; S143, According to road condition type The preset road condition-feature adaptation rule table is used to assign adaptation weights to various features, resulting in a road condition adaptation weight set for each feature. , Represented as the first Adaptation weights for class feature assignment; S144. Calculate the initial weights: ； In the formula, Indicates the first Initial weights for class features; Simultaneously, by normalizing the fusion weights, the sum of the fusion weights of all features is ensured to be 1, resulting in a normalized fusion weight set. ; S145. Multiply each classification feature by its corresponding normalized fusion weight and sum them to obtain the fusion feature: ； In the formula, Indicates the first Class fusion features, These represent the indices for environmental fusion features, driver fusion features, and vehicle fusion features, respectively. Indicates the first The subset of original features corresponding to the class fusion features; S146, Integration of Environmental Characteristics Driver fusion characteristics and the characteristics of vehicle integration This forms the final fused sensing data. .

6. The intelligent driving method applicable to multiple road conditions according to claim 5, characterized in that: Step S2 specifically includes the following steps: S21. Mean oxyhemoglobin characteristics in the fused sensing data based on the output of S1 Characteristics of mean deoxyhemoglobin and power spectral density characteristics Input the JKNN-SVM model, output the driver state classification results, and then calculate the risk value by combining the state duration; S211, Mean characteristics of paraoxyhemoglobin Characteristics of mean deoxyhemoglobin and power spectral density characteristics Z-score standardization is performed to obtain the standardized feature vector. ; S212. The similarity between the test features and all samples in the training set is calculated using Jaccard distance, and the top-ranked features are then selected. One nearest neighbor sample; S213, the output of S212 The nearest neighbor sample set optimizes the training sample distribution of the SVM, and the RBF kernel function is used to construct the classifier. The penalty coefficient and kernel function parameters of the SVM are determined through Bayesian parameter optimization, and the driver state level is output. : ； in, ； In the formula, Indicates the SVM decision score; Represents the standardized feature vector to be classified; Represent the Lagrange multipliers of SVM; The labels representing the training samples are the driver's state levels. Represents the kernel function, and , This represents the width parameter of the RBF kernel function. This represents the feature vector of a sample in the training set; Indicates the bias term; Indicates the preceding Total number of nearest neighbor samples; S214. Driver status level obtained based on classification and state duration Quantifying driver distraction risk using the Sigmoid function : ； In the formula, Indicates the risk growth coefficient; Indicates the threshold for the duration of the state; S22. Quantify environmental visibility risks separately. Road surface friction risk and collision risk : ； ； ； In the formula, This indicates the time of collision between the vehicle and the obstacle in front, and , Indicates the relative distance between the vehicle and the obstacle; Indicates the collision warning threshold; S23. Construct a coupled risk model based on Bayesian networks to address driver distraction risk. Environmental visibility risk Road surface friction risk Collision risk As an input node, total risk As the output node, the coupling relationship between various risk factors is quantified through a conditional probability table to obtain the total risk. and risk level ; in, ; In the formula, Indicates that given network parameters Under these conditions, driver distraction risk Visibility risk Road surface friction risk Collision risk The probability of them happening simultaneously; Indicates the parameters of the Bayesian network; Indicates; in the Parent node of each risk factor Status and network parameters Next, the The probability of each risk factor occurring These respectively represent the risk of driver distraction. Environmental visibility risk Road surface friction risk Collision risk index.

7. The intelligent driving method applicable to multiple road conditions according to claim 6, characterized in that: Step S3 specifically includes the following steps: S31. Normalize the road surface friction coefficient Available lane distance Speed ​​deviation Risk level Road curvature As indicators, weights are dynamically assigned to each indicator according to a pre-defined scenario-weight mapping table. The normalized indicators and the assigned dynamic weights are then substituted into the comprehensive evaluation function to calculate the sum score for each candidate lane. : ； In the formula, , , , and These represent the normalized road surface friction coefficient, lane availability distance, vehicle speed deviation, risk level, and road curvature, respectively. , , , , These are respectively represented as the normalized road surface friction coefficient, lane availability distance, vehicle speed deviation, risk level, and dynamic weights of road curvature distribution; S32, Sum score of all candidate lanes Sort the lanes in descending order, exclude lanes that are not allowed to proceed, and select the lane with the highest score as the optimal decision. ; S33. Extract shallow spatial features and deep semantic features using the MAFPN fusion step S13: ； In the formula, This represents the final fused feature output by the improved YOLOv10 multi-scale flexible convolutional module; S34. Predicting the future using extended Kalman filtering. obstacle movement trajectory within seconds This provides dynamic obstacle constraints for trajectory planning; among them, express The location of obstacles at all times. express The speed of the obstacle at any given moment; S35. Use fifth-order polynomial programming for the lateral trajectory: ； In the formula, express The vehicle's lateral position at all times; , , , , and All represent the coefficients of a fifth-order polynomial; Represents a time variable; By utilizing boundary constraints and introducing lateral acceleration constraints under multiple road conditions, the coefficients of the fifth-order polynomial are solved: Boundary constraints: ; Lateral acceleration constraints under multiple road conditions: ; In the formula, Indicates the initial lateral position of the vehicle; Indicates the current lateral position of the vehicle; Indicates the horizontal position of the target; This indicates the target lateral position of the vehicle at the moment the trajectory ends; This represents the lateral velocity of the vehicle at the initial moment; This represents the lateral velocity of the vehicle at the moment the trajectory ends; This represents the lateral acceleration of the vehicle at the initial moment; This represents the lateral acceleration of the vehicle at the moment the trajectory ends; express Horizontal velocity at all times; Indicates the maximum lateral acceleration; Represents gravitational acceleration; get , , , , , ; S36. Use a cubic acceleration model to plan the longitudinal trajectory, combined with the risk level. and the trajectory of the obstacle The longitudinal acceleration is dynamically adjusted to match the longitudinal velocity with the lateral trajectory. The expression for the cubic acceleration model is as follows: ； In the formula, Indicates the future The longitudinal speed of the vehicle at all times; This indicates the current longitudinal speed of the vehicle; Indicates the initial longitudinal acceleration; Indicates the rate of change of longitudinal acceleration; Constraints: Speed ​​constraints: ; Acceleration constraints: ; Collision avoidance constraints: ; In the formula, Indicates the minimum safe speed; express longitudinal acceleration at all times ; Indicates the maximum longitudinal acceleration; express The longitudinal position of the vehicle at all times; Indicates the length of the vehicle; express The longitudinal position of the obstacle ahead at any given moment; Indicates a safe distance; S37. Construct a longitudinally and laterally coupled frame, and adjust the longitudinal velocity and lateral acceleration through coupling constraints: ； ； Coupling constraints: ; In the formula, This indicates the longitudinal velocity after coupling adjustment; This represents the lateral acceleration after coupling adjustment; This indicates the lateral acceleration before adjustment. ; S38. Model predictive control is used to smooth and optimize the coupled longitudinal and lateral trajectories, with trajectory smoothness, safety, and efficiency as optimization objectives. The optimization weights are dynamically adjusted to adapt to multiple road conditions to obtain the initial trajectory of the vehicle. : ； Among them, the model predictive control optimization objective function The expression is as follows: ； Constraints: Vertical constraints: and ; Lateral constraints: and ; Safety constraints: ； In the formula, , , , Indicates the optimization weights; Represents the optimal decision The optimal speed in the system; Indicates the minimum distance between the vehicle and the obstacle; These represent the lateral positions of the left and right boundaries of the lane, respectively. S39. With trajectory smoothness, safety, and energy efficiency as optimization objectives, the initial trajectory is iteratively optimized using the MPC algorithm. Output the optimal trajectory , Indicates the optimal position. Indicates the optimal speed. This represents the optimal longitudinal acceleration. This represents the optimal lateral acceleration. Indicates the optimal heading angle; The multi-objective optimization function is expressed as follows: ； Constraints: ; In the formula, , and Both represent optimized weights; Indicates the real-time distance between the vehicle and obstacles in front; S310, From the optimal trajectory Extract the target deceleration at the current moment. Deceleration duration and speed difference : ； ； ； In the formula, Indicates the current vehicle speed; express The target vehicle speed after time step is determined by the optimal trajectory. Obtain; Indicates the default deceleration time domain; This indicates the maximum permissible deceleration, and , Indicates the safety redundancy coefficient; S311, Based on road surface friction coefficient Total Risk Value The optimal regenerative braking efficiency is calculated by adjusting for the vehicle's remaining battery power. : ； in, ； In the formula, Indicates the basic regeneration efficiency; Indicates the preset regenerative braking target efficiency; , and All represent weighting coefficients; S312, Based on total braking intensity and optimal regenerative braking efficiency Distribute hydraulic braking intensity and regenerative braking strength : ； ； in, ； Road condition adaptation constraints: when hour, and ; when hour, and ; when hour, and ; In the formula, Represents the optimal trajectory The target deceleration; This represents the corrected deceleration time domain; S313, targeting On icy and snowy roads, a closed-loop slip ratio control system is introduced to control wheel slip ratio. The optimal range is 10%-15%, which can be achieved through adjustments. and The dynamic response speed prevents wheel lock-up or skidding. ； ； in, ； ； In the formula, and These represent the hydraulic braking strength and regenerative braking strength after slip ratio correction, respectively; This represents the slip ratio correction factor; Indicates real-time wheel speed; Indicates the rolling radius of the wheel; S314, in conjunction with risk level Adjust the response priority of the braking strategy; when When it is high risk, , ; when When it is medium risk, , ; when When it is low risk, , ; In the formula, and These represent the final hydraulic braking strength and regenerative braking strength after risk adaptation, respectively. S315, Output Braking Strategy , This indicates the preset target slip ratio.

8. The intelligent driving method applicable to multiple road conditions according to claim 7, characterized in that: Step S4 Specifically, the following steps are included: S41, the optimal trajectory output by S3 Risk assessment results of S2 , The dataset is obtained by performing timestamp alignment and noise suppression on the sensing data of S1. ; in, These represent the optimal trajectory, total risk, driver status level, road friction coefficient, and vehicle status after synchronous preprocessing, respectively. S42. Aiming at the optimal trajectory after synchronization, based on the vehicle dynamics model and future... The deviation of each prediction step is used to construct the MPC objective function, and the pre-heading angle is generated under road condition constraints. and pre-throttle opening ; Wherein, the MPC objective function The expression is as follows: ； Control constraints: ,and ; Dynamic constraints: ; In the formula, , , , , , All represent deviation weights; , , , and These represent the prediction steps. The longitudinal position deviation, lateral position deviation, longitudinal velocity deviation, longitudinal acceleration deviation, and heading angle deviation of the vehicle at any given time; Indicates control over the time domain; Indicates the prediction time domain; and Both represent the smoothing weight of the control quantity; express Increment of heading angle at any moment; express The increment of throttle opening at any given moment; and These represent the maximum and minimum heading angles, respectively. express Heading angle at any moment; express Throttle opening at all times; express lateral acceleration at all times; S43. Utilize dual PID controllers to track lateral position and longitudinal velocity, where the lateral PID controller uses... and Input: , Output: Final heading angle Vertical PID and For output, output the final valve opening. ; The PID control expression for the lateral position is as follows: ； in, ； In the formula, , and These represent the proportional, integral, and derivative parameters of the lateral position PID control, respectively. express lateral positional deviation at any moment; and They represent Time and The horizontal integral term at time step; express Rate of change of position deviation at time; Indicates the integral separation threshold; Indicates the sampling time interval; The longitudinal velocity PID control expression is as follows: ； in, ； In the formula, , and These represent the proportional, integral, and derivative parameters of the longitudinal velocity PID control, respectively. express Momentary velocity deviation; and They represent Time and The longitudinal integral term at time step; express Rate of change of velocity deviation at any given time; and These represent the upper and lower limits of saturation for vertical integration, respectively; Represents; represents a saturation function; S44. After integrating and limiting the braking signal and trajectory tracking signal, output the final steering, acceleration and braking signals; Among them, steering control: Acceleration control: ; Braking control: ; And during the driving process, based on the total risk after pretreatment and driver status level Output the following early warning strategy: when or When it enters the mild warning mode, it will output a voice warning. when or When the system enters a severe warning mode, it will output seat vibration and voice warning. when When entering emergency warning mode, it will be forcibly activated. Upgraded to The longitudinal speed decreased to ; express The optimal longitudinal speed of the vehicle at any given time is taken from the optimal speed. ; Furthermore, when entering emergency mode, braking control is set to the highest priority.

Images