Vehicle-road collaborative awareness and fusion estimation vehicle stability control method facing beyond visual range scene

Abstract

The invention discloses a vehicle-road collaborative perception and fusion estimation vehicle stability control method for a beyond-visual-range scene, and belongs to the technical field of crossing of intelligent traffic, vehicle control engineering and an automobile safety technology. The method comprises the following steps: acquiring vehicle dynamic data through a vehicle end sensor, estimating a dynamic pavement adhesion coefficient based on a seven-degree-of-freedom model and unscented Kalman filtering, acquiring a pavement image by using a roadside camera, identifying a pavement state by using a MobileNetV3 + model, and mapping the pavement state into a visual pavement adhesion coefficient; a discrete time Kalman filter is adopted to fuse dynamics and visual estimation results, and the fusion weight is dynamically adjusted according to the confidence coefficient; and transmitting the fused road adhesion coefficient to a vehicle end controller, and designing an upper-layer controller and a lower-layer controller based on sliding mode control and road adhesion coefficient identification, thereby realizing stability control of the vehicle under complex road conditions. The accuracy and foresight of road adhesion coefficient estimation can be effectively improved, and the handling stability of the vehicle under the limit working condition is enhanced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent transportation, vehicle control engineering and automobile safety technology, and particularly relates to a vehicle-road cooperative perception and fusion estimation vehicle stability control method for an over-the-horizon scene. BACKGROUND

[0002] In a traditional vehicle stability control system, the road adhesion coefficient is usually estimated by relying on vehicle sensors or vehicle end cameras, and control decisions are made based on the same. However, these methods still have limitations.

[0003] The vehicle sensors are difficult to perceive the road conditions in front of the vehicle in advance, especially in a visual blind area or a complex road condition, and the warning and control capabilities are limited. Moreover, there is a certain lag in the estimation of dynamics, which depends on the current state of the vehicle, and feedforward control cannot be achieved.

[0004] The traditional control method lacks a cooperative mechanism and does not fully utilize the information interaction between the roadside facilities and the vehicle, and the overall advantage of the "vehicle-road-cloud" cooperation cannot be brought into play. The road adhesion coefficient of the rear vehicle cannot be estimated more accurately and more prospectively through the fusion of multiple sources of information such as roadside cameras, front vehicle dynamics data, and cloud data processing, so as to improve the handling stability of the whole vehicle in extreme conditions.

[0005] Therefore, there is an urgent need in the prior art for a new technical solution to solve the above problems. SUMMARY

[0006] The technical problem to be solved by the present application is to provide a vehicle-road cooperative perception and fusion estimation vehicle stability control method for an over-the-horizon scene, so as to solve the problems of control lag and failure of the vehicle in complex road environments, especially in a visual blind area or an accident-prone curve, due to the inability of the vehicle to timely and accurately obtain the road adhesion coefficient in front of the vehicle.

[0007] The vehicle-road cooperative perception and fusion estimation vehicle stability control method for an over-the-horizon scene is triggered in a visual blind area and a specific curved road section determined by accident history data; the following steps are included and performed in sequence. Step 1: Obtain the vehicle dynamics road adhesion coefficient. Step 2: Obtain the visual road adhesion coefficient of the road end camera. Step 3: Fuse the visual road adhesion coefficient and the dynamics road adhesion coefficient. Step 4: Establish communication between the road end workstation and the vehicle end. Step 5: Judge the current instability risk based on the road adhesion coefficient. Step 6: Execute the vehicle handling stability control strategy according to the judgment of the risk.

[0008] The vehicle dynamics adhesion coefficient calculation method of step one is: collecting vehicle dynamics data in driving, establishing a seven-degree-of-freedom model of the vehicle, obtaining longitudinal, lateral and yaw motion equations, four wheel vertical loads, side slip angles and slip rates; through the Dugoff tire model, obtaining the normalized tire force, and using the unscented Kalman filter algorithm to obtain the estimated dynamics road adhesion coefficient μ dyn .

[0009] The visual road surface adhesion coefficient calculation method of the road end camera of step two is: collecting vehicle front road surface image data, and performing data preprocessing; inputting the preprocessed data into a lightweight convolutional neural network model MobileNetV3+, using deep separable convolution and bottleneck structure to extract multi-level features, and outputting the road surface state category through the classifier; establishing a mapping relationship lookup table between the predefined road surface state category and the adhesion coefficient value, querying and outputting the corresponding estimated visual road surface adhesion coefficient μ vis , and evaluating the visual confidence C of this visual estimation through the probability vector output by the Softmax layer of the MobileNetV3+ model vis .

[0010] The fusion visual road surface adhesion coefficient and dynamics road surface adhesion coefficient method of step three is: Step 1: using a discrete-time Kalman filter with state as adhesion coefficient μ, establishing a system model: , In the formula, w k is the process noise, which is subject to Gaussian distribution with mean of zero and variance of Q; Step 2: substituting two observation sources: estimated dynamics road adhesion coefficient μ dyn and estimated visual road surface adhesion coefficient μ vis ; , In the formula, and are observation noises, and are subject to Gaussian distribution with mean of zero and variance of R , Step 3: dynamically adjusting observation noise variances and according to confidence Dynamics observation noise variance: ; In the formula, is the confidence of the dynamics estimate, S is the vehicle state, where, is a computation function representing the dynamics observation noise variance, specifically the base noise variance , the inverse of the dynamics confidence , and an adjustment factor based on the vehicle state are multiplied together; is specifically represented as: , is the visual observation noise variance: , where: : base noise variance of visual observation, representing the inherent uncertainty of visual estimate under ideal conditions; : confidence of visual estimate, output by visual recognition model; : adjustment factor based on environmental conditions E, including light intensity, weather condition, road texture, camera occlusion, and factors affecting visual recognition accuracy; : adjustment function based on road variation flag , when = 1, indicating that a road type change is detected, the function reflects higher uncertainty by increasing the noise variance; E: environmental conditions, including light, weather, road texture, and visual perception related factors; Step 4: Confidence calculation, dynamics confidence and visual confidence are calculated by the following formula: , where, and are confidence calculation functions; Step 5: Road variation detection detects road type changes through visual observation value mutation: , When exceeds the threshold and is large, it is determined that the road has changed, triggering flag 1; Step 6: Kalman filtering and final fusion results: Prediction step: , Update step: ① Calculate Kalman gain: , wherein, ; ②State update: , wherein, ; ③Covariance update: , The state output of the filter is the fused estimate of the road adhesion coefficient.

[0011] The method for the road-side station to establish communication with the vehicle-side is as follows: The road-side unit sends a data packet to the vehicle-side unit, including the fused estimate of the road adhesion coefficient, the road segment position information corresponding to the fused estimate of the road adhesion coefficient, the data validity timestamp, and the confidence level of the road-side unit for the coefficient; The road-side unit includes a camera and a road-side computing station, the camera is configured to collect road surface images, and the road-side computing station is configured to perform the following operations: identify the visual road adhesion coefficient based on the images and fuse it with the dynamic road adhesion coefficient received from the vehicle; The vehicle-side unit includes a vehicle-mounted sensor, a communication module, and a vehicle controller, the vehicle-mounted sensor is configured to estimate the dynamic road adhesion coefficient, the communication module is configured to interact with the road-side unit, and the vehicle controller is configured to receive the fused road adhesion coefficient and execute the vehicle stability control strategy.

[0012] A cloud server is further included, which is in communication connection with the road-side unit and the vehicle-side unit, and is configured to store and distribute historical road surface data and fusion results.

[0013] Step five: judging the current instability risk based on the road adhesion coefficient, including wheel lock, side slip caused by centrifugal force, and excessive steering caused by the vehicle under certain driving conditions.

[0014] Step six: executing the vehicle stability control strategy according to the judgment of the risk, including the design of an upper controller based on sliding mode control and a lower torque distribution method based on road adhesion coefficient identification.

[0015] The lower torque distribution method based on road adhesion coefficient identification is as follows: Let the total yaw moment T = T1 + T2+ T3, wherein: T1 = W1 ( F x -F xd ) 2 , T2 = W2 ( M z -M zd) 2 , T3 = W3 [ ( F x11 D x ) 112 + ( F x12 D x ) 122 + ( F x21 D x ) 212 + ( F x22 D x ) 222 ], F x =F x11 cosδ 11 +F x12 cosδ 12 +F x21 cosδ 21 +F x22 cosδ 22 , M z = F x11 h1+F x12 h2+F x21 h3+F x22 h4, where: T: total yaw moment generated by the distribution of wheel torques; T1: weighted square term of longitudinal force tracking error, representing the deviation of actual total longitudinal force F x from the desired total longitudinal force F xd , used for optimizing longitudinal force tracking; T2: weighted square term of yaw moment tracking error, representing the deviation of actual yaw moment M z from the desired yaw moment M zd , used for optimizing yaw moment tracking; T3: weighted square sum of wheel adhesion utilization, representing the square sum of the ratio of each wheel's longitudinal force F to the maximum longitudinal force F , used for optimizing tire adhesion margin to prevent wheel slip or lock; M z : actual yaw moment calculated from each wheel's longitudinal force F and its force arm h1 to the vehicle center; M zd : desired yaw moment calculated by the upper-layer sliding mode controller based on vehicle state and road adhesion coefficient; F x : actual total longitudinal force calculated from each wheel's longitudinal force F and its steering angle The cosine projection of the longitudinal force of each wheel to the vehicle longitudinal axis is summed up to get the total longitudinal force F xd : The desired total longitudinal force, calculated by the upper controller according to the driver input and vehicle state δ 11 : The steering angle of the front left wheel (the first wheel), used to calculate the longitudinal force projection and yaw moment W1: The longitudinal force weight coefficient W2: The weight coefficient of the yaw moment W3: The weight coefficient of the four-wheel adhesion coefficient hi: The effective distance from the longitudinal force of each wheel to the center of the vehicle D xij : The maximum longitudinal force that the ground can provide to each wheel, the solution formula is: D xij = μ xij F zij , Where: μ xij is the maximum friction factor between the road and the wheel calculated according to the PAC tire model; F zij is the vertical load of each wheel The local optimal solution is obtained by the Lagrange multiplier method, assuming that the objective function is a convex function formula, that is, the Hessian matrix is positive definite; The Hessian matrix H of the objective function is expressed as: , Let = 0, we can get: , Where: : The steering angle of each wheel L i : The distance from the center of mass to each axle A: The cosine matrix of the steering angle, the element , is used to project the longitudinal force of each wheel to the vehicle longitudinal axis to calculate the total longitudinal force ; B: The yaw moment contribution matrix, the element , where is the distance from the center of mass to each axle, j represents the left and right positions of the wheel, and is used to calculate the contribution of the longitudinal force of each wheel to the yaw moment ; D: The adhesion utilization rate weight matrix, which is a diagonal matrix, the element is , where is the maximum longitudinal force that the ground can provide to each wheel, used to weight the utilization of the adhesion of each wheel in the objective function; and are elements in matrix A and B respectively.

[0016] Through the above design scheme, the vehicle-road cooperative perception and fusion estimation vehicle stability control method for over-the-horizon scenarios can solve the problems of control lag and failure of vehicle stability control caused by the vehicle's inability to timely and accurately obtain the road adhesion coefficient in front of the vehicle in complex road environments, especially in visual blind areas or accident-prone curves, effectively improving the accuracy and forward-looking of the road adhesion coefficient estimation, and enhancing the vehicle's handling stability in extreme working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0017] The application will be further described below in combination with the drawings and specific embodiments: Figure 1 is a schematic diagram of the application scenario.

[0018] Figure 2 is a schematic diagram of the dynamics estimation road adhesion coefficient of the application.

[0019] Figure 3 is a schematic diagram of the vision estimation road adhesion coefficient of the application.

[0020] Figure 4 is a schematic diagram of the fusion of dynamics and vision estimation road adhesion coefficient process of the application.

[0021] Figure 5 is a schematic diagram of the vehicle-road cooperative perception and fusion estimation vehicle stability control method for over-the-horizon scenarios of the application.

[0022] Figure 6 is a schematic diagram of the communication between the road end workstation and the vehicle end workstation of the application. DETAILED DESCRIPTION

[0023] The vehicle-road cooperative perception and fusion estimation vehicle stability control method for over-the-horizon scenarios, as shown in Figures 1-6 , includes the following steps: Step one, obtaining the vehicle dynamics adhesion coefficient.

[0024] Using an accelerometer, a gyroscope sensor to collect the dynamics data of the vehicle during driving, including the acceleration, speed, steering angle, lateral acceleration, yaw rate, wheel speed, and mass center side slip angle of the vehicle; The collected data is preprocessed by removing noise and filtering; A seven-degree-of-freedom model of the vehicle is established to obtain the longitudinal, lateral and yaw motion equations of the vehicle, as well as the calculation formulas of the vertical load, side slip angle and slip ratio of the four wheels. The longitudinal motion equation of the vehicle is: , The lateral motion equation of the vehicle is: , The longitudinal and yaw motion equation of the vehicle is as follows: , In the formula, the subscripts fl, fr, rl and rr in the letters respectively represent the front left wheel, the front right wheel, the rear left wheel and the rear right wheel of the vehicle, is the longitudinal vehicle speed, is the lateral vehicle speed, x_ij is the wheel longitudinal force, y_ij is the wheel lateral force, z is the moment of inertia about the Z axis, is the left front wheel angle, is the right front wheel angle, is the yaw angle, is the front wheel track, is the rear wheel track, is the distance from the front axle to the center of mass, is the distance from the rear axle to the center of mass.

[0025] Wheel vertical load: , In the formula, are the tire longitudinal forces of the front left wheel, the front right wheel, the rear left wheel and the rear right wheel, is the tire weight, is the vehicle body weight, L is the wheelbase, is the vehicle center of mass height, is the longitudinal acceleration of the vehicle, is the lateral acceleration of the vehicle.

[0026] Tire side slip angle: , In the formula, are the tire side slip angles of the front left wheel, the front right wheel, the rear left wheel and the rear right wheel, are the longitudinal speeds of the front left wheel, the front right wheel, the rear left wheel and the rear right wheel, are the lateral speeds of the front left wheel, the front right wheel, the rear left wheel and the rear right wheel.

[0027] Tire slip ratio: , The Dugoff tire model is established to obtain the tire normalized tire force; , In the formula, μ represents the road adhesion coefficient, F z is the vertical load, λ is the slip ratio, α is the tire side slip angle, ε is the speed influence factor, C x is the longitudinal stiffness of each tire, C y is the side slip stiffness of each tire.

[0028] The formula for tire normalization is as follows: , In the formula, , represent the tire normalized longitudinal force and normalized lateral force.

[0029] The road adhesion coefficient μ 1。

[0030] The state variable is set as , the measurement variable is , and the control variable u(t) = [δ]; the corresponding prediction equation is: , The observation equation is: , Where:

[0031] Step 2: Obtain the visual road adhesion coefficient of the roadside camera.

[0032] As a roadside sensing device, the roadside camera collects and analyzes the image of the road ahead to accurately identify the road condition and obtain the visual road adhesion coefficient estimate. The specific steps are as follows: (1) Roadside image acquisition and preprocessing: Deploy high-definition cameras at key locations on the road (such as curves, hilltops, and accident-prone sections) to continuously collect image data of the road ahead at a preset sampling frequency. Preprocess the collected raw images, including grayscale, contrast enhancement, Gaussian filtering to suppress noise, and divide the region of interest (ROI) to focus on the effective road surface area on the vehicle's driving track.

[0033] (2) Pavement condition recognition based on MobileNetV3+ model: A lightweight convolutional neural network model MobileNetV3+ is used as the core classifier. The pre-processed pavement image is input into the model, and its deep separable convolution and bottleneck structure are used to efficiently extract multi-level features, and finally the classifier outputs the pavement condition category. The category is based on a pre-defined classification system, including but not limited to dry asphalt pavement - light damage, dry asphalt pavement - heavy damage, wet asphalt pavement - light damage, wet asphalt pavement - heavy damage, snow-covered pavement, icy pavement, etc.

[0034] (3) Visual pavement friction coefficient mapping: a mapping relationship lookup table between a pre-defined pavement condition category and a typical friction coefficient value is established. According to the specific pavement condition category identified by the MobileNetV3+ model in step (2), the corresponding nominal visual pavement friction coefficient estimate value is queried and output from the lookup table . For example, when identified as "icy pavement", the mapped value is 0.15; when identified as "dry asphalt pavement - light damage", the mapped value is 0.85.

[0035] (4) Visual estimation confidence evaluation: at the same time, the confidence of this visual estimation is evaluated according to the probability vector output by the final Softmax layer of the MobileNetV3+ model . The confidence calculation formula is: , where P i is the probability value of the model determining the pavement image as the i-th category. The higher the value, the higher the credibility of the visual recognition result, and the confidence will be used for dynamic adjustment of the fusion weight in step three.

[0036] (5) Data output: the road-side computing unit sends the calculated visual pavement friction coefficient estimate value and its corresponding confidence to the relevant vehicle through the vehicle-road cooperative communication link described in step four.

[0037] Step three, fusion of visual pavement friction coefficient and dynamic pavement friction coefficient.

[0038] Step 1: Use a discrete-time Kalman filter with state as friction coefficient μ to establish a system model: , where w k is the process noise, which follows a Gaussian distribution with mean zero and variance Q; Step 2: Substitute two observation sources: estimated dynamic road adhesion coefficient μ dyn and estimated visual road adhesion coefficient μ vis ; , wherein, and are observation noises, and respectively follow Gaussian distribution with mean zero and variance σ2, and are independent of each other, the covariance matrix of observation noises is: , Step 3: Dynamically adjust observation noise variance according to confidence and ; Dynamic observation noise variance: ; wherein, is the confidence of dynamic estimation, S is the vehicle state, wherein, represents the calculation function of dynamic observation noise variance, which is specifically obtained by multiplying the basic noise variance , the inverse of the dynamic confidence , and the adjustment factor based on the vehicle state; Specifically, it is expressed as: , Visual observation noise variance: , wherein: : the basic noise variance of visual observation, representing the inherent uncertainty of visual estimation under ideal conditions; : the confidence of visual estimation, output by the visual recognition model; : the adjustment factor based on the environmental condition E, including light intensity, weather condition, road texture, camera obstruction, and factors affecting visual recognition accuracy; : the adjustment function based on the road change sign , when = 1, it means that the road type change is detected, and the function reflects higher uncertainty by increasing the noise variance; E: environmental conditions, including light, weather, road texture, and visual perception related factors; Step 4: Confidence calculation, dynamic confidence and visual confidence The following formula is used for calculation: , In the formula, and is a confidence calculation function; Step 5: Road surface change detection detects road surface type changes through visual observation value mutation: , When exceeds the threshold value and is large, it is determined that the road surface changes, triggering flag 1; Step 6: Kalman filtering and final fusion of results: Prediction step: , Update step: ① Calculate the Kalman gain: , In the formula, ; ② State update: , In the formula, ; ③ Covariance update: , The final fusion result, the state output of the filter is the fused adhesion coefficient estimate.

[0039] Step four, the road end workstation and the vehicle end establish communication contact.

[0040] The intelligent network system can mainly be divided into "vehicle-road-cloud" three information modules, through the information interaction of all parties, the vehicle, road, and cloud end are coordinated, and finally the road adhesion coefficient estimate is transmitted.

[0041] The data packet sent by the road end unit to the vehicle end unit at least includes: the fused road adhesion coefficient value, the road section position information corresponding to the coefficient, the data effective timestamp, and the confidence level of the road end to the coefficient.

[0042] A road side unit, comprising at least one camera and a road end computing workstation, the camera is used to collect road surface images, and the road end computing workstation is configured to perform the following operations: identify visual road adhesion coefficient based on image recognition, and fuse with the dynamic road adhesion coefficient received from the vehicle.

[0043] The vehicle end unit comprises a vehicle-mounted sensor for estimating a dynamic road adhesion coefficient, a communication module for data interaction with the roadside unit, and a vehicle controller configured to receive the fused road adhesion coefficient and execute a handling stability control strategy based thereon.

[0044] The cloud server is in communication connection with the roadside unit and the vehicle end unit, and is used for storing and distributing historical road data and fusion results.

[0045] Step five: judging the current instability risk based on the road adhesion coefficient.

[0046] (1) Wheel lock. According to the single wheel model, in order to prevent wheel lock, the brake force of the brake must be less than the friction force provided by the ground to the wheel: , In order to prevent wheel lock, the brake force of the brake must be less than the friction force provided by the ground to the wheel.

[0047] (2) Side slip caused by centrifugal force. When the vehicle moves at a speed V x on a circle with a radius r, once the centrifugal force is greater than the lateral force F y provided by the tire, the vehicle will lose stability.

[0048] (3) Excessive steering caused by the vehicle in some driving conditions. Vehicles with insufficient steering have a certain high-speed stability, but generally, vehicles will have a tendency to deflect to the inside when braking on a curve. The reason for this tendency is that the load of the rear wheel is shifted to the front wheel to a large extent when the vehicle brakes, so that the positive moment is greater than the negative moment.

[0049] Step six: executing a vehicle handling stability control strategy according to the judgment of the risk.

[0050] (1) Design of longitudinal vehicle speed sliding mode controller based on sliding mode control. According to the theory of vehicle dynamics, the calculation formula of the total longitudinal force of the vehicle ∑ Fx can be obtained as: , In the formula: is the longitudinal acceleration of the vehicle; γ is the yaw rate; v is the lateral vehicle speed; and m is the mass of the vehicle.

[0051] The rate of change of the sliding mode surface of the longitudinal vehicle speed sliding mode controller can be expressed by a function with a saturation function as: , In the formula: is the sliding mode surface of the longitudinal vehicle speed sliding mode controller, H v = u -u dWhere u is the actual longitudinal vehicle speed, u d For the desired longitudinal speed; sat( H v ) is a saturation function, and , In the formula: v represents the boundary layer thickness designed for the longitudinal vehicle speed sliding mode controller; C v ε v This is the convergence law coefficient.

[0052] The desired longitudinal resultant force ∑ F is obtained xd The expression is: ∑ F xd = m[-ε v sat( H v -C v H v -γ v + u(·) d ], In the formula: u(·) d The desired longitudinal acceleration of the vehicle.

[0053] Yaw angular velocity sliding mode controller According to vehicle dynamics theory, the total yaw force of the vehicle can be obtained. The expression for ∑ M is: ∑ M = I z γ In the formula I z Let H(·) be the moment of inertia about the Z-axis at the vehicle's center of mass. The yaw rate is the rate of change of the sliding surface of the sliding mode controller. γ It can be expressed as a function with a saturation function as follows: H(·) γ = -ε γ sat( H γ -k γ H γ In the formula H γ H is the sliding surface for the yaw rate sliding mode controller. γ = γ -γ d Where γ is the actual yaw rate, γ d For the desired yaw rate; sat( H γ ) is a saturation function, and , In the formula: k γ ε γ C is the coefficient of the reaching law; sat γ is the boundary layer thickness designed for the yaw rate sliding mode controller.

[0054] The expression for the desired total yaw moment ∑ M is: ∑ M = I z [ -ε γ sat( H γ ) -k γ H γ +γ(·) d ] where: γ(·) d is the desired yaw rate change.

[0055] Lower layer torque distribution method based on road adhesion coefficient identification T = T1 + T2 + T3 where: T1 = W1 ( F x -F xd ) 2 , T2 = W2 ( M z -M zd ) 2 , T3 = W3 [ ( F x11 D x ) 112 + ( F x12 D x ) 122 + ( F x21 D x ) 212 + ( F x22 D x ) 222 ], F x =F x11 cosδ 11 +F x12 cosδ 12 +F x21 cosδ 21 +F x22 cosδ 22 , M z = F x11 h1+F x12 h2+F x21 h3+F x22 h4, where: T is the total yaw moment; W1, W2, W3 are weight coefficients of longitudinal force, yaw moment, and four-wheel adhesion coefficients, respectively; F xij is the longitudinal force of each wheel that needs to be distributed; hi is the effective distance from the longitudinal force of each wheel to the center of the vehicle under force; D xijThe maximum longitudinal force that the ground can provide to each wheel is calculated by the formula: D xij = μ xij F zij where μ xij is the maximum friction coefficient between the road and the wheel calculated according to the PAC tire model; F zij is the vertical load of each wheel.

[0056] The solution obtained by the Lagrange multiplier method is a local optimum. To obtain a global optimum, it is necessary to assume that the objective function is a convex function, i.e., the Hessian matrix is positive definite. The Hessian matrix H of the objective function is expressed as: , Let = 0, and the following is obtained: , where is the steering angle of each wheel; L i is the distance from the center of mass to each axle.

Claims

1. A vehicle-road cooperative perception and fusion estimation vehicle stability control method for over-the-horizon scenes, triggered at specific curved road sections determined by visual blind areas and accident history data; characterized in that: Comprising the following steps, And the following steps are sequentially carried out: Step one: obtain the vehicle dynamics road adhesion coefficient; Step two: obtain the visual road adhesion coefficient of the road end camera; Step three: fuse the visual road adhesion coefficient and the dynamics road adhesion coefficient; Step four: the road end workstation establishes communication contact with the vehicle end; Step five: judge the current instability risk based on the road adhesion coefficient; Step six: execute the vehicle handling stability control strategy according to the risk judgment.

2. The method of claim 1, wherein the method is used for over-the-horizon scene vehicle-road cooperative perception and fusion estimation vehicle stability control. The vehicle dynamics adhesion coefficient calculation method in step one is: collecting vehicle dynamics data in driving, establishing a seven-degree-of-freedom model of the vehicle, obtaining longitudinal, lateral and yaw motion equations, four wheel vertical loads, side slip angles and slip rates; through the Dugoff tire model, obtaining the normalized tire force, and using the unscented Kalman filter algorithm to obtain the estimated dynamics road adhesion coefficient μ dyn .

3. The method of claim 1, wherein the method further comprises: The method for calculating the visual road surface adhesion coefficient of the road end camera in step two is: collecting road surface image data in front of the vehicle, performing data preprocessing; inputting the preprocessed data into a lightweight convolutional neural network model MobileNetV3+, extracting multi-level features by using depth separable convolution and bottleneck structure, and outputting the road surface state category through a classifier; establishing a mapping relationship lookup table between the predefined road surface state category and the adhesion coefficient value, and querying and outputting the corresponding estimated visual road surface adhesion coefficient μ vis The visual confidence C vis of the visual estimation is evaluated through the probability vector output by the final Softmax layer of the MobileNetV3+ model.

4. The method of claim 1, wherein the method further comprises: The method for fusing the visual road adhesion coefficient and the dynamics road adhesion coefficient in step three is: Step 1: use a discrete-time Kalman filter with state as adhesion coefficient μ to establish system model: ， where w k is process noise, subject to a Gaussian distribution with zero mean and variance Q; Step 2: Substitute in two observation sources: estimated dynamic road adhesion coefficient μ dyn and estimated visual road adhesion coefficient μ vis ; ， where and is the observation noise, σ2 dynand σ2 visrespectively follow Gaussian distributions with zero mean and variance σ2 dynand σ2 vis, respectively, and are independent of each other, and the covariance matrix of the observation noise is: ， Step 3: dynamically adjust observation noise variance σ2 dyn and σ2 vis according to confidence level; Dynamics observation noise variance: ； wherein is a confidence of the dynamics estimate, S is a vehicle state, wherein represents a computation function of the dynamics observation noise variance, specifically obtained by multiplying the inverse of the base noise variance , the inverse of the dynamics confidence , and an adjustment factor based on the vehicle state ; Specifically expressed as: ， Visual observation noise variance: ， In the formula: : the variance of the noise underlying visual observations, representing the inherent uncertainty of visual estimates under ideal conditions; : confidence of visual estimation, output by visual recognition model; : adjustment factor based on environmental conditions E, including light intensity, weather conditions, road texture, camera obstruction, and factors affecting visual recognition accuracy; Based on road surface change signs The adjustment function, when When =1, it indicates that a change in road surface type has been detected, and the function reflects higher uncertainty by increasing the noise variance; E: environmental conditions, including light, weather, road texture and visual perception related factors; Step 4: Confidence calculation, Kinetic Confidence and Visual Confidence calculated by the following equation: , In the formula, and is a confidence calculation function; Step 5: road change detection detects road type change through sudden change of visual observation value: , When exceeds the threshold and is large, it is determined that the road surface changes, and flag 1 is triggered ​ Step 6: Kalman filtering and final fusion result: Prediction step: ， Update step: ① Calculate Kalman gain: ， In the formulae, ; ② State update: ， In the formulae, ; ③ Covariance update: ， State output of filter is the fused estimate of the attachment coefficient.

5. The method of claim 1, wherein the method further comprises: The method for the road end workstation to establish communication contact with the vehicle end in step four is: The road end unit sends data packets to the vehicle end unit, including the fused adhesion coefficient estimate value, the road section position information corresponding to the fused adhesion coefficient estimate value, the data effective timestamp, and the confidence level of the road end to the coefficient; The road side unit includes a camera and a road end computing workstation, and the camera is used to collect road images; The road end computing workstation is configured to perform the following operations: identify the visual road adhesion coefficient based on the image, and fuse it with the dynamics road adhesion coefficient received from the vehicle; The vehicle end unit includes a vehicle-mounted sensor, a communication module and a vehicle controller, the vehicle-mounted sensor is used to estimate the dynamics road adhesion coefficient, the communication module is used for data interaction with the road side unit, and the vehicle controller is configured to receive the fused road adhesion coefficient and execute the handling stability control strategy.

6. The method of claim 5, wherein the method further comprises: It also includes a cloud server in communication connection with the road side unit and the vehicle end unit, used for storing and distributing historical road data and fusion results.

7. The method of claim 1, wherein the method further comprises: The judgment of current instability risk based on the road adhesion coefficient in step five includes wheel lock, centrifugal force induced side slip and excessive steering caused by the vehicle under certain driving conditions.

8. The method of claim 1, wherein the method further comprises: determining a target speed of the ego vehicle based on the target speed of the ego vehicle and the target speed of the target vehicle. The execution of the vehicle handling stability control strategy according to the risk judgment in step six includes the design of the upper controller based on sliding mode control and the lower torque distribution method based on road adhesion coefficient identification.

9. The method of claim 8, wherein the method further comprises: The lower torque distribution method based on road adhesion coefficient identification is: Let total yaw moment T = T1 + T2+ T3, where: T1 = W1 ( F x - F xd ) 2 , T2 = W2 ( M z -M zd ) 2 , T3 = W3 [ ( F x11 D x ) 112 + ( F x12 D x ) 122 + ( F x21 D x ) 212 + ( F x22 D x ) 222 ], F x =F x11 cosδ 11 +F x12 cosδ 12 +F x21 cosδ 21 +F x22 cosδ 22 , M z = F x11 h1+F x12 h2+F x21 h3+F x22 h4, In the formula: T: total yaw moment, generated by each wheel torque distribution; T1 weighted square term of the longitudinal force tracking error, representing the deviation of the actual total longitudinal force F x from the desired total longitudinal force F xd for optimizing the longitudinal force tracking; T2: weighted square term of yaw moment tracking error, representing deviation of actual yaw moment M z from desired yaw moment M zd for optimizing yaw moment tracking; T3: Weighted sum of square of each wheel adhesion utilization, representing longitudinal force of each wheel to the maximum longitudinal force ratio, for optimizing tire adhesion margin and preventing wheel slip or lock M z : actual yaw moment, calculated from the longitudinal forces of the individual wheels and their lever arms hi to the center of the vehicle ; M zd : desired yaw moment, calculated by the upper layer sliding mode controller according to vehicle states and road adhesion coefficient; F x : actual total longitudinal force, summed from the longitudinal forces of the individual wheels and the cosine of their steering angle projected onto the longitudinal vehicle axis F xd : desired total longitudinal force, calculated by the upper controller from driver input and vehicle state; δ 11 : steering angle of the front left wheel, used to calculate the longitudinal force projection and the yaw moment; W1: longitudinal force weight coefficient; W2: yaw moment weight coefficient; W3: four-wheel adhesion coefficient weight coefficient; hi: the effective distance from each wheel longitudinal force to the center of force of each vehicle; D xij : The maximum longitudinal force that the ground can provide to each wheel is calculated using the following formula: D xij = μ xij F zij , where: μ xij is the maximum friction factor between the road and the wheel calculated according to the PAC tire model; F zij is the vertical load of each wheel; The local optimal solution is obtained by the Lagrange multiplier method, assuming that the objective function is a convex function, i.e. the Hessian matrix is positive definite; the Hessian matrix H of the objective function is expressed as: ， Let = 0, we have: ， In the formula: : wheel angle; L i : centroid-to-axis distance; A: Cosine matrix of the steering angle, elements , for projecting the individual wheel longitudinal forces into the vehicle longitudinal axis direction, calculating the total longitudinal force ; B: yaw moment contribution matrix, element where is the distance from the center of mass to each axle, j indicates the left or right position of the wheel, and is used to calculate the contribution of each wheel's longitudinal force to the yaw moment ​ D: the adhesion utilization rate weight matrix, is a diagonal matrix, and the elements are wherein is the maximum longitudinal force that the ground can provide to each wheel, used to weight the adhesion utilization rate of each wheel in the objective function; and are elements in matrices A and B, respectively.

Images