Lateral control method and device based on road feature recognition

By sensing road features in real time and dynamically matching lateral control parameters, the problem of control stability and trajectory tracking accuracy of intelligent driving systems under changing road conditions is solved, achieving more stable vehicle driving control.

CN123104660APending Publication Date: 2026-11-17CHERY AUTOMOBILE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611450443.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-17
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing intelligent driving lateral control systems struggle to adapt to complex road features in real time when road conditions change dynamically, leading to control gain mismatch, response lag, and steering oscillations, which affect trajectory tracking accuracy and driving smoothness.

Method used

By sensing road surface images in real time, visual features such as road surface type, adhesion coefficient, roughness, and texture direction are extracted, and lateral control parameters are dynamically matched and input into the lateral controller for vehicle control, thereby achieving real-time parameter adaptation.

Benefits of technology

It improves the control consistency and ride comfort of the vehicle under different road conditions, reduces gain mismatch and control oscillation caused by fixed or lagging parameters, and enhances trajectory tracking accuracy and driving smoothness.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123104660A_ABST
    Figure CN123104660A_ABST
Patent Text Reader

Abstract

The lateral control method and apparatus based on road surface feature recognition provided in this application include: real-time acquisition of road surface images; multi-source feature perception of the obtained road surface images to obtain visual road surface features, wherein the visual road surface features include: road surface type, estimated road surface adhesion coefficient, road surface roughness, and road surface texture direction; matching corresponding lateral control parameters in control parameter mapping data according to the visual road surface features; inputting the lateral control parameters into a lateral controller, and performing lateral control of the vehicle through the lateral controller; thereby solving the technical problem in related technologies of how to improve the adaptability of lateral control parameters to complex road surface features under dynamic changes in road surface conditions, while taking into account both control stability and ride comfort.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, and in particular to a lateral control method and device based on road feature recognition. Background Technology

[0002] With the rapid development of intelligent driving technology, the autonomous driving capabilities of vehicles on structured roads have become increasingly mature, and functions such as Lane Centering Control (LCC) and Navigation-Assisted Driving (NOA) have been gradually mass-produced and implemented in vehicles. However, in actual road driving, the diversity of road conditions poses a severe challenge to the control performance of intelligent driving systems. Existing intelligent driving lateral control systems generally adopt lane centering and path tracking control based on vehicle speed segment compensation or offline calibration parameters. However, these solutions lack real-time correlation with physical characteristics such as road surface adhesion coefficient, roughness, and texture direction. Parameter adjustments are mainly based on fixed operating condition settings. When switching between wet, rough, or structurally textured road surfaces, control gain mismatch, response lag, and steering oscillations are prone to occur, affecting trajectory tracking accuracy and driving smoothness.

[0003] Therefore, it is urgent to solve the technical problem of how to improve the adaptability of lateral control parameters to complex road features under dynamic changes in road conditions, while taking into account both control stability and ride comfort. Summary of the Invention

[0004] This application provides a lateral control method and apparatus based on road surface feature recognition to solve the aforementioned technical problems. This solution addresses the vehicle's lateral control process by sensing relevant road surface features in real time and associating the obtained road surface features with preset control parameters. This allows the lateral control parameters to dynamically adapt to the current road surface state, thereby improving the consistency and adaptability of vehicle control when performing lane centering and path tracking under different road surface conditions.

[0005] In a first aspect, this application provides a lateral control method based on road surface feature recognition, comprising:

[0006] Real-time acquisition of road surface images, multi-source feature perception of the obtained road surface images to obtain visual road surface features, including: road surface type, estimated road surface adhesion coefficient, road surface roughness, and road surface texture direction.

[0007] Match the corresponding lateral control parameters in the control parameter mapping data based on visual road surface features;

[0008] The lateral control parameters are input into the lateral controller, which then controls the vehicle laterally.

[0009] In one possible implementation, the obtained road surface image is subjected to multi-source feature perception to obtain visual road surface features, including:

[0010] Road surface features are extracted from the road surface image to obtain road surface image features, which include: road surface type, road surface roughness, and road surface texture direction;

[0011] The road surface adhesion coefficient is estimated based on the road surface image features, and the estimated value of the road surface adhesion coefficient is obtained.

[0012] In one possible implementation, after performing multi-source feature perception on the obtained road surface image to obtain visual road surface features, the method further includes:

[0013] The confidence level of visual road surface features is evaluated based on the moving road surface features, which are calculated based on vehicle kinematic parameters.

[0014] When the confidence level is greater than the first threshold, the lateral control parameters are matched in the control parameter mapping data based on the visual road surface features.

[0015] In one possible implementation, matching lateral control parameters in the control parameter mapping data based on visual road surface features includes:

[0016] The index key is determined based on visual road surface features and the vehicle's real-time speed.

[0017] Multidimensional linear interpolation is performed on the control parameter mapping data based on the index key to match the lateral control parameters, which include: proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency.

[0018] In one possible implementation, after matching lateral control parameters in the control parameter mapping data based on visual road surface features, the method further includes:

[0019] The feedforward gain is dynamically adjusted based on the direction of the road surface texture.

[0020] In one possible implementation, lateral control parameters are input to a lateral controller, and the lateral controller performs lateral control of the vehicle, including:

[0021] The transverse control parameters are smoothed by a first-order low-pass filter to obtain the processed transverse control parameters.

[0022] The processed lateral control parameters are input into the lateral controller, which then performs lateral control on the vehicle.

[0023] In one possible implementation, the transverse control parameters are smoothed using a first-order low-pass filter, including:

[0024] Real-time calculation of the feature change rate of visual road surface features;

[0025] The smoothing coefficient is dynamically adjusted based on the characteristic rate of change.

[0026] The transverse control parameters are smoothed using a first-order low-pass filter based on the adjusted smoothing coefficient.

[0027] In one possible implementation, road surface features are extracted from the road surface image to obtain road surface image features, including:

[0028] Optical flow analysis is performed on consecutive frames of road surface images to extract the motion vector field of road surface texture points;

[0029] Calculate the angle between the main direction of the texture and the longitudinal axis of the vehicle's movement;

[0030] The direction of road surface texture is identified based on the included angle feature, where the direction of road surface texture includes: longitudinal texture, transverse texture and non-directional texture.

[0031] In one possible implementation, road surface features are extracted from the road surface image to obtain road surface image features, including:

[0032] Gray-level co-occurrence matrices are constructed from multiple directions, and texture contrast features in multiple directions are calculated.

[0033] Calculate the mean of multiple texture contrast features to obtain the contrast parameters;

[0034] The contrast parameter is mapped to road surface roughness through a pre-calibration relationship.

[0035] Secondly, this application provides a lateral control device based on road surface feature recognition, comprising:

[0036] The feature perception module is used to acquire road surface images in real time, perform multi-source feature perception on the obtained road surface images, and obtain visual road surface features, including: road surface type, estimated road surface adhesion coefficient, road surface roughness, and road surface texture direction.

[0037] The matching module is used to match the corresponding lateral control parameters in the control parameter mapping data based on the visual road surface features;

[0038] The lateral control module is used to input lateral control parameters into the lateral controller, which then performs lateral control on the vehicle.

[0039] In one possible implementation, the feature-aware module is further used for:

[0040] Road surface features are extracted from the road surface image to obtain road surface image features, which include: road surface type, road surface roughness, and road surface texture direction;

[0041] The road surface adhesion coefficient is estimated based on the road surface image features, and the estimated value of the road surface adhesion coefficient is obtained.

[0042] In one possible implementation, the feature-aware module is further used for:

[0043] The confidence level of visual road surface features is evaluated based on the moving road surface features, which are calculated based on vehicle kinematic parameters.

[0044] When the confidence level is greater than the first threshold, the lateral control parameters are matched in the control parameter mapping data based on the visual road surface features.

[0045] In one possible implementation, the matching module is also used for:

[0046] The index key is determined based on visual road surface features and the vehicle's real-time speed.

[0047] Multidimensional linear interpolation is performed on the control parameter mapping data based on the index key to match the lateral control parameters, which include: proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency.

[0048] In one possible implementation, the matching module is also used for:

[0049] The feedforward gain is dynamically adjusted based on the direction of the road surface texture.

[0050] In one possible implementation, the lateral control module is also used for:

[0051] The transverse control parameters are smoothed by a first-order low-pass filter to obtain the processed transverse control parameters.

[0052] The processed lateral control parameters are input into the lateral controller, which then performs lateral control on the vehicle.

[0053] In one possible implementation, the lateral control module is also used for:

[0054] Real-time calculation of the feature change rate of visual road surface features;

[0055] The smoothing coefficient is dynamically adjusted based on the characteristic rate of change.

[0056] The transverse control parameters are smoothed using a first-order low-pass filter based on the adjusted smoothing coefficient.

[0057] In one possible implementation, the feature-aware module is further used for:

[0058] Optical flow analysis is performed on consecutive frames of road surface images to extract the motion vector field of road surface texture points;

[0059] Calculate the angle between the main direction of the texture and the longitudinal axis of the vehicle's movement;

[0060] The direction of road surface texture is identified based on the included angle feature, where the direction of road surface texture includes: longitudinal texture, transverse texture and non-directional texture.

[0061] In one possible implementation, the feature-aware module is further used for:

[0062] Gray-level co-occurrence matrices are constructed from multiple directions, and texture contrast features in multiple directions are calculated.

[0063] Calculate the mean of multiple texture contrast features to obtain the contrast parameters;

[0064] The contrast parameter is mapped to road surface roughness through a pre-calibration relationship.

[0065] Thirdly, this application provides an adapter device for a lateral control method based on road feature recognition, including: a memory and a processor;

[0066] The memory stores instructions that the computer executes;

[0067] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0068] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible embodiments of the first aspect.

[0069] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0070] This application provides a lateral control method and device based on road surface feature recognition. By acquiring road surface images in real time and extracting visual road surface features such as road surface type, adhesion coefficient estimate, roughness, and texture direction, it is possible to establish a correspondence between lateral control parameters and the current physical state of the road surface. This allows for timely matching of more suitable control parameters under conditions of wet, rough, or significantly textured road surfaces. By inputting the matched lateral control parameters into the lateral controller, it is possible to reduce gain mismatch and control oscillations caused by fixed or lagging parameters, thereby improving the stability of vehicle lateral control, trajectory tracking accuracy, and driving smoothness. Attached Figure Description

[0071] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0072] Figure 1 This application provides a schematic diagram of an application data processing system architecture.

[0073] Figure 2 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 1 ;

[0074] Figure 3 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 2 ;

[0075] Figure 4 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 3 ;

[0076] Figure 5 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 4 ;

[0077] Figure 6 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 5 ;

[0078] Figure 7 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 6 ;

[0079] Figure 8 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 7 ;

[0080] Figure 9 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 8 ;

[0081] Figure 10 A flowchart of the adaptive method for lateral control parameters based on road feature recognition provided in the embodiments of this application;

[0082] Figure 11 This is a flowchart illustrating the online road feature recognition process provided in an embodiment of this application.

[0083] Figure 12 A schematic diagram of the structure of the road surface feature-control parameter mapping table provided in the embodiments of this application;

[0084] Figure 13 Comparison chart of the smooth transition effect of control parameters provided in the embodiments of this application;

[0085] Figure 14 This application provides a system hardware architecture block diagram for an adaptive lateral control parameter system based on road feature recognition, as shown in the embodiments of this application.

[0086] Figure 15 A schematic diagram of the structure of the lateral control device based on road feature recognition provided in the embodiments of this application;

[0087] Figure 16 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0088] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0089] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0090] In the field of lateral control technology for intelligent driving vehicles, L2 to L4 level lane centering and navigation-assisted driving functions are often deployed in structured and open road environments. During vehicle operation, vehicles frequently encounter conditions such as dry road surfaces, slippery sections, rough paving, bridge expansion joints, and local potholes. The lateral control system needs to continuously adjust steering commands based on road conditions to maintain stable vehicle movement along the target trajectory.

[0091] Existing lateral control schemes typically rely on segmented vehicle speed compensation or offline calibration parameters. Some schemes estimate road conditions through visual recognition or vehicle dynamics, then map the results to the controller gain. However, these schemes have limited real-time correlation capabilities with road surface adhesion coefficient, roughness, and texture direction. Control parameters are often pre-set around fixed operating conditions, making them prone to gain mismatch, response lag, and steering oscillations when road conditions change abruptly. Single feature inputs are also susceptible to interference from changes in illumination, sensor noise, and transient slippage, resulting in unstable road surface recognition results, which in turn affects trajectory tracking accuracy and ride comfort.

[0092] Based on the aforementioned technical challenges, vehicle lateral control requires first acquiring visual road surface features that can reflect road type, adhesion state, surface undulations, and texture direction. Then, lateral control parameters are matched and input into the lateral controller based on these features. Following this approach, this application employs an onboard road surface image acquisition and multi-source feature perception link to continuously identify and match road surface conditions. This allows control parameters to adjust synchronously with changes in road surface physical characteristics, reducing control deviations caused by fixed parameters during transitions to complex road surfaces and improving the stability and ride comfort of lateral control.

[0093] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0094] Figure 1 This is a schematic diagram of an application data processing system architecture provided in an embodiment of this application. The application data processing system is a computer device. Figure 1 As shown, the above architecture includes at least one of a data acquisition device 101, a processing device 102, and a display device 103.

[0095] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the architecture of the application data processing system. In other feasible embodiments of this application, the above architecture may include more or fewer components than illustrated, or combine some components, or split some components, or arrange different components, which can be determined according to the actual application scenario and is not limited here. Figure 1 The components shown can be implemented in hardware, software, or a combination of both.

[0096] In the specific implementation process, the data acquisition device 101 may include an input / output interface or a communication interface, and the data acquisition device 101 can be connected to the processing device through the input / output interface or the communication interface.

[0097] The processing device 102 can acquire road surface images in real time, perform multi-source feature perception on the obtained road surface images, and obtain visual road surface features, including road surface type, estimated road surface adhesion coefficient, road surface roughness, and road surface texture direction; match the corresponding lateral control parameters in the control parameter mapping data according to the visual road surface features; input the lateral control parameters into the lateral controller, and perform lateral control of the vehicle through the lateral controller.

[0098] The display device 103 can also be a touch screen or the screen of a terminal device, used to receive user commands while displaying the above-mentioned content, so as to realize interaction with the user.

[0099] It should be understood that the aforementioned processing device can be implemented by a processor reading instructions from memory and executing those instructions, or it can be implemented by a chip circuit.

[0100] Furthermore, the network architecture and business scenarios described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0101] Figure 2 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 1 ,like Figure 2 As shown, the lateral control method based on road feature recognition provided in this embodiment includes:

[0102] S201. Real-time acquisition of road surface images, multi-source feature perception of the obtained road surface images to obtain visual road surface features, including road surface type, road surface adhesion coefficient estimate, road surface roughness, and road surface texture direction.

[0103] In one possible embodiment, road surface image acquisition is performed by a vehicle-mounted forward-facing camera, which is installed at the front of the vehicle and oriented towards the direction of travel. This camera continuously acquires images of the road ahead while the vehicle is in motion. The acquisition process is performed at a preset sampling frequency to form a temporally continuous sequence of road surface images, ensuring that each frame corresponds to the current state of the road surface ahead. After acquisition, the image data is fed into a multi-source feature perception process, which may include preprocessing and feature extraction to obtain visual road surface features for subsequent control. The road surface type can be identified based on regional features in the road surface image. Road surface roughness and texture direction can be obtained by analyzing the texture information of the road surface region. The road surface adhesion coefficient estimate is calculated by combining road surface image features with real-time vehicle kinematic parameters to characterize the frictional adhesion state of the tire-road contact interface. By incorporating road surface type, road surface adhesion coefficient estimate, road surface roughness, and road surface texture direction into the visual road surface features, subsequent control parameter matching can simultaneously consider road surface appearance attributes and adhesion state information, and output a complete feature vector for the next step.

[0104] S202. Match the corresponding lateral control parameters in the control parameter mapping data based on the visual road surface characteristics.

[0105] In one possible embodiment, control parameter mapping data is used to establish the correspondence between road surface features and lateral control parameters, and is read by the controller during the control process. Lateral control parameters characterize the control gain configuration of the lateral controller under the current road surface conditions, and include one or more of proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency. The control parameter mapping data uses road surface type, estimated road adhesion coefficient, road surface roughness, and road surface texture direction as search conditions, corresponding to different parameter combinations. For example, when a wet road surface is identified and the estimated adhesion coefficient is low, the mapping data can output a lower proportional gain, a lower feedforward gain, and a lower filter cutoff frequency, so that the controller can generate a smoother control quantity under low adhesion conditions; when a dry road surface with low roughness is identified, the mapping data can output another set of parameter combinations. This matching process can update the control parameters in a timely manner when the road surface condition changes, and output a set of parameters adapted to the current road surface condition to the lateral controller.

[0106] S203. Input the lateral control parameters into the lateral controller, and use the lateral controller to perform lateral control on the vehicle.

[0107] In one possible embodiment, the lateral controller receives lateral control parameters output from control parameter mapping data and uses them in the vehicle's lateral control closed loop. The lateral controller generates control commands based on the vehicle's current state and the target trajectory, which are then converted into front wheel steering angle or steering torque via the steering actuator. The lateral controller can employ a PID structure, a feedforward plus feedback structure, or a combined control structure including filtering, where the proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency act on the error term, error variation term, trajectory feedforward term, and high-frequency noise suppression term, respectively. During the control process, the controller updates the control input and sends it to the steering actuator, causing the vehicle to travel along the target lane centerline or navigation reference trajectory. Because the lateral control parameters are updated based on real-time road surface characteristics, the controller can use different parameter combinations for lateral correction when traversing different road surfaces, maintaining a correspondence between the control commands and the current road surface state, thus enabling continuous execution of the lateral control process and completing vehicle path tracking.

[0108] This application provides a lateral control method based on road surface feature recognition. By acquiring road surface images in real time and performing multi-source feature perception, the method unifies road surface type, estimated road surface adhesion coefficient, road surface roughness, and road surface texture direction into visual road surface features. Then, based on these features, lateral control parameters are matched in control parameter mapping data and input into the lateral controller, enabling the control parameters to be updated synchronously with the physical state of the road surface. This method can achieve online adaptation to changes in road surface state and directly use the matching results for vehicle lateral control. It should be understood that the above example is merely illustrative and not limiting. This application can also employ other implementation methods capable of achieving road surface image acquisition, multi-source feature perception, control parameter mapping, and lateral control, as long as they achieve the same or equivalent technical effects without departing from the technical concept of this application, they can all be considered exemplary implementations of this application.

[0109] In one possible implementation, Figure 3 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 2 ,like Figure 3 As shown, multi-source feature perception is performed on the obtained road surface image to obtain visual road surface features, including:

[0110] S301, Extract road features from the road surface image to obtain road surface image features, which include: road surface type, road surface roughness, and road surface texture direction;

[0111] S302, the road surface adhesion coefficient is estimated based on the road surface image features to obtain the estimated value of the road surface adhesion coefficient.

[0112] Among them, road surface image features are used to carry the main road surface information extracted from the road surface image and serve as input for the adhesion coefficient estimation; road surface type is used to characterize the current road surface category; road surface roughness is used to characterize the surface undulation and particle distribution; and road surface texture direction is used to characterize the spatial extension trend of the road surface texture. Together, these three constitute a visual description of the current road surface condition. The estimated road surface adhesion coefficient is used to characterize the adhesion ability between the tire and the road surface and provides a basis for subsequent lateral control parameter matching.

[0113] In practical implementation, road surface images can be continuously acquired by an onboard camera within the vehicle's forward field of view. After denoising, distortion correction, and road surface region segmentation, the acquired images are input into a feature extraction module. This module can be implemented using a convolutional neural network, texture operators, and semantic segmentation networks. The convolutional network is used to identify road surface types such as asphalt, concrete, and wet pavement; the texture operator is used to extract local grayscale fluctuations to characterize roughness; and directional filtering or gradient statistics are used to determine the dominant texture direction. For adhesion coefficient estimation, the system can select at least one of the following implementation methods based on the onboard platform's computing power and sensor configuration: Implementation method one: Match the road surface type, road surface roughness, and texture direction with a pre-established adhesion coefficient mapping relationship, or input the data into a regression model trained based on historical data to calculate the estimated road adhesion coefficient; Implementation method two: Simultaneously acquire vehicle kinematic parameters, and use wheel speed difference, lateral acceleration, and yaw rate as inputs to the tire dynamics model or kinematic formula for inverse calculation to obtain the estimated road adhesion coefficient. The correlation between the estimated adhesion coefficient and the current road surface condition is clearer, and the lateral control parameters can be adjusted according to changes in the road surface, thereby improving the control stability and trajectory tracking consistency during lane centering.

[0114] This method simultaneously extracts road surface type, roughness, and texture direction, and estimates the adhesion coefficient accordingly. This allows visual road surface features to cover the correlation between road surface condition and tire operating conditions, enabling subsequent control parameter matching to be based on a more complete road surface description. Using this method, the correspondence between the estimated adhesion coefficient and the current road surface condition is clearer, and lateral control parameters can be adjusted according to road surface changes, thereby improving control stability and trajectory tracking consistency during lane centering.

[0115] In an optional embodiment, Figure 4 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 3 ,like Figure 4 As shown, after performing multi-source feature perception on the obtained road surface image to obtain visual road surface features, the method further includes:

[0116] S401, assess the confidence level of visual road surface features based on moving road surface features, wherein the moving road surface features are calculated based on vehicle kinematic parameters;

[0117] S402, when the confidence level is greater than the first threshold, the lateral control parameters are matched in the control parameter mapping data according to the visual road surface features.

[0118] In this embodiment, visual road surface features are used to characterize the current road surface state, and can be composed of road surface type, estimated road surface adhesion coefficient, road surface roughness, and road surface texture direction extracted from the road surface image. Vehicle kinematic parameters are used to reflect the dynamic state of the vehicle during the current driving process, and can specifically consist of parameters such as wheel speed, acceleration, and yaw rate, which are collected or calculated in real time within the vehicle controller. The motion road surface features are derived from the above-mentioned vehicle kinematic parameters, used to characterize the dynamic response features related to the road surface state, and serve as the basis for consistency verification of visual road surface features.

[0119] After acquiring visual road surface features, the controller performs correlation analysis between these features and moving road surface features, generating a confidence score based on the degree of consistency, correlation, or matching between the two. This confidence score can be calculated using a predefined evaluation function, which maps the deviations between the visual and moving road surface features in terms of adhesion changes, undulation disturbances, or texture responses into numerical results. A higher value indicates a higher level of confidence in the visual road surface features. The first threshold is a preset threshold value used to limit the minimum confidence condition for the visual road surface features to enter the subsequent control parameter matching stage. The threshold value can be pre-stored in the controller and can be set based on the vehicle platform, sensor configuration, or road test results.

[0120] When the confidence level exceeds a first threshold, the controller searches or indexes the control parameter mapping data based on the visual road surface features to obtain the corresponding lateral control parameters. The control parameter mapping data can be stored as a correspondence table between visual road surface features and lateral control parameters, where different combinations of road surface types, adhesion coefficient estimates, roughness, and texture direction correspond to different steering gains, feedforward coefficients, or lateral control corrections. Lateral control parameters can be preset in memory as table entries, and the processor reads and outputs them according to the feature values ​​of the current visual road surface features. In practical applications, this storage structure can also be implemented using arrays, hash maps, or lookup tables; this application does not limit its implementation in this regard.

[0121] In this manner, visual road surface features are first validated for reliability using moving road surface features before entering the control parameter mapping data matching process. This ensures that lateral control parameters are updated only when the features are reliable. Since vehicle kinematic parameters can provide dynamic evidence of changes in road surface conditions, this confidence-gating method makes the matching results more stable and maintains a more consistent correspondence between lateral control parameters and the current road surface condition, thereby improving the continuity of lateral control and the stability of trajectory tracking.

[0122] In one possible embodiment, the confidence level of visual road surface features is evaluated based on the motion road surface characteristics, specifically including a fusion evaluation process based on DS evidence theory. The system uses the visual recognition results obtained from the extraction of road surface image features as the first evidence body, and the motion road surface features calculated based on vehicle kinematic parameters as the second evidence body. First, the two evidence bodies are normalized respectively, and the basic probability assignments of each proposition (road surface type, adhesion coefficient estimate, roughness, and texture direction) are calculated. Then, the first and second evidence bodies are subjected to an orthogonality test, and the conflict coefficient is calculated. When the conflict coefficient is lower than a preset upper limit, a weighted fusion is performed according to the DS synthesis rules to generate a fused joint basic probability assignment. The system calculates the confidence level corresponding to the maximum probability value in the joint basic probability assignment. If the confidence level is greater than a first threshold of 0.6, the visual road surface features are determined to be credible, and subsequent parameter matching steps are executed; if the confidence level is less than or equal to the first threshold of 0.6, the features are determined to have conflict or uncertainty, the current visual feature output is frozen, and the lateral control parameters of the previous cycle remain unchanged until the data of the new cycle meets the confidence level condition.

[0123] In one possible embodiment, Figure 5 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 4 ,like Figure 5 As shown, lateral control parameters are matched in the control parameter mapping data based on visual road surface features, including:

[0124] S501, determine the index key based on visual road surface features and the vehicle's real-time speed;

[0125] S502 performs multidimensional linear interpolation operations on the control parameter mapping data based on the index key to match the lateral control parameters, which include: proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency.

[0126] The index key is used to locate the matching position in the control parameter mapping data. It is constructed by combining visual road surface features and the vehicle's real-time speed, and serves as the joint input for interpolation retrieval. The vehicle's real-time speed is the speed value collected at the current moment, which participates in the construction of the index key and affects the matching result of the lateral control parameters. Multidimensional linear interpolation is used to obtain continuous parameter values ​​between multiple adjacent mapping points, so that the proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency are smoothly updated with changes in road surface conditions and vehicle speed.

[0127] In the specific implementation, the control parameter mapping data can be pre-constructed as a multi-dimensional working condition lookup table structure. The table contains a discrete calibration node grid consisting of typical road surface types, adhesion coefficient ranges, roughness thresholds, texture direction categories, and vehicle speed ranges. After the vehicle acquires road surface images and obtains visual road surface features during operation, these features are combined with the real-time vehicle speed to form an index key corresponding to the current working condition. The system locates the discrete node group adjacent to the index key in the lookup table structure, performs multi-dimensional linear interpolation operations on the continuous feature dimension, performs category matching and weight allocation on the category feature dimension, and fuses the interpolation results with the category weights. The proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency are then weighted and solved to output the lateral control parameters continuously corresponding to the current working condition. The low-pass filter cutoff frequency is used to limit the passage range of high-frequency components, the feedforward gain is used to compensate for lateral control, and the proportional gain and derivative gain are used to adjust the response intensity of the deviation and its rate of change, respectively.

[0128] This method integrates visual road surface features and vehicle speed into parameter localization, and obtains continuous control quantities through multidimensional linear interpolation. This allows lateral control parameters to be adjusted synchronously with changes in road conditions and speed. Consequently, the controller input parameters maintain a smooth transition, reducing parameter abrupt changes caused by interval switching, and improving the stability of lateral control and the consistency of trajectory tracking.

[0129] In one possible implementation, after matching lateral control parameters in the control parameter mapping data based on visual road surface features, the method further includes: dynamically adjusting the feedforward gain according to the road surface texture direction.

[0130] Among them, the road surface texture direction is used to characterize the spatial extension direction of the road surface texture. It can be represented by the texture gradient direction, stripe direction, or texture principal direction vector in the road surface image, and can serve as the directional basis for the correction of lateral control parameters. The feedforward gain is used to adjust the amplitude of the feedforward compensation channel in the lateral controller. The output of the feedforward compensation channel is used to pre-compensate for the control requirements caused by road geometry and road surface conditions before the steering control quantity is formed.

[0131] In its implementation, after completing the lateral control parameter matching, the controller reads the road texture direction corresponding to the current moment and compares this direction with a preset direction threshold or direction category to determine the correction coefficient of the feedforward gain. When the road texture direction is basically consistent with the vehicle's driving direction, the feedforward gain can be corrected according to the corresponding down-adjustment coefficient; when the road texture direction is laterally distributed with the vehicle's driving direction, the feedforward gain can maintain the mapped parameter value or be corrected according to a preset compensation coefficient; when the road texture direction has no obvious directionality, the correction amount is determined by combining the road roughness and road adhesion coefficient estimates, and the feedforward gain is updated accordingly. The correction process can be pre-stored as a lookup table relationship in the control parameter mapping data, or a secondary correction term can be superimposed on the matched feedforward gain. Both can be implemented through real-time calculation by the on-board controller. In practical applications, the controller can use an embedded processor to complete the feedforward gain update, and the relevant parameter table can be stored in non-volatile memory; this application does not limit this.

[0132] In this way, the feedforward gain can be adjusted synchronously with the change of road surface texture direction, so that the feedforward compensation intensity matches the directional characteristics of the road surface texture, thereby keeping the adaptability of the lateral control parameters consistent on different textured road surfaces and making the steering output more stable and continuous.

[0133] In one possible implementation, Figure 6 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 5 ,like Figure 6 As shown, lateral control parameters are input into the lateral controller, which then performs lateral control of the vehicle, including:

[0134] S601 uses a first-order low-pass filter to smooth the lateral control parameters, resulting in the processed lateral control parameters.

[0135] S602 inputs the processed lateral control parameters into the lateral controller, which then performs lateral control on the vehicle.

[0136] The first-order low-pass filter is used to continuously process the updates of the lateral control parameters, filtering out high-frequency variation components caused by abrupt parameter changes, thus ensuring that the lateral control parameters remain numerically smooth and continuous. The processed lateral control parameters are the filtered control quantities, which can correspond to all or some of the parameters in the lateral controller, including the proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency. The processed lateral control parameters can be directly written into the parameter register or control interface of the lateral controller at output, and can be retrieved by the controller within the current control cycle.

[0137] In practical implementation, the first-order low-pass filter can perform continuous transition calculations using a discrete recursive approach. The output parameters at the current moment are generated by combining the filtered output from the previous moment with the current input parameters, and the filtering time constant is set according to the controller's sampling period. The lateral controller can be a steering controller based on lane centering or navigation-assisted driving. After receiving the processed parameters, it combines the vehicle's lateral deviation, heading angle deviation, and vehicle speed information to output steering control values ​​and drive the vehicle to perform lateral correction. The filter model and controller model used can be configured according to the computing power and interface type of the vehicle controller. In practical applications, other models can also be selected for this component, and this application does not limit this.

[0138] By smoothing the lateral control parameters before inputting them into the lateral controller, the control parameters can remain continuous during the update process, reducing the direct impact of step changes in parameters on the control output and making the lateral control process more stable, thereby improving the smoothness and control consistency of the vehicle when traveling along the target trajectory.

[0139] In one possible embodiment, Figure 7 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 6 ,like Figure 7 As shown, the transverse control parameters are smoothed using a first-order low-pass filter, including:

[0140] S701 calculates the feature change rate of visual road surface features in real time and dynamically adjusts the smoothing coefficient based on the feature change rate.

[0141] S702 smooths the transverse control parameters using a first-order low-pass filter based on the adjusted smoothing coefficient.

[0142] Visual road surface features refer to the road surface state representation obtained from road surface images through multi-source feature perception, including at least road surface type, estimated road surface adhesion coefficient, road surface roughness, and road surface texture direction. The feature change rate characterizes the speed of change of visual road surface features within adjacent periods and can be calculated from the feature difference amplitude, normalized difference component, or weighted change amount at adjacent sampling times. The smoothing coefficient adjusts the attenuation strength of the first-order low-pass filter for the lateral control parameters. Increasing the smoothing coefficient speeds up the follow-up speed of the filter output to new input parameters, while decreasing the smoothing coefficient results in a smoother filter output.

[0143] In its implementation, the controller receives the current visual road surface features in each control cycle, compares the current cycle features with those of the previous cycle, calculates the corresponding feature change, and then combines this with the cycle time interval to obtain the feature change rate. This feature change rate can be obtained by summing the differences of individual features, or by assigning weights to road surface type, adhesion coefficient estimate, roughness, and texture direction for comprehensive calculation, reflecting the overall trend of road surface conditions. Subsequently, the system generates a smoothing coefficient based on the feature change rate and a preset mapping relationship. When the change rate is large, the smoothing coefficient is increased to accelerate the convergence of the lateral control parameters; when the change rate is small, the smoothing coefficient is decreased to suppress parameter jitter. The first-order low-pass filter can use a discrete recursive form to update the proportional gain, derivative gain, feedforward gain, or low-pass filter cutoff frequency, and the filtered output serves as the smoothed parameters input to the lateral controller. The mapping relationship can be stored in a control parameter adjustment table. In practical applications, this table can also be configured according to vehicle type, speed range, and sensor sampling frequency; this application does not impose any limitations on this.

[0144] Based on the above processing, the system can incorporate the dynamic changes in visual road surface features into the smoothing parameter adjustment, enabling the lateral control parameters to maintain a strong response when the road surface condition changes rapidly and a high degree of stability when the road surface condition is stable. This makes the parameter sequence input to the lateral controller more continuous. As a result, the fluctuation of the control command of the lateral controller is reduced in transitional road surface scenarios such as wet, rough, bridge gaps, or potholes, parameter mismatch and steering oscillation are suppressed, and the vehicle's lateral tracking stability and ride comfort are correspondingly improved.

[0145] Optionally, Figure 8 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 7 ,like Figure 8 As shown, road surface features are extracted from the road surface image to obtain the road surface image features, including:

[0146] S801 performs optical flow analysis on consecutive frames of road surface images to extract the motion vector field of road surface texture points;

[0147] S802, calculates the angle feature between the main direction of the texture and the longitudinal axis of the vehicle's movement;

[0148] S803 identifies the direction of road surface texture based on the included angle feature, where the direction of road surface texture includes: longitudinal texture, transverse texture and non-directional texture.

[0149] Among them, continuous frame images refer to adjacent image sequences used for temporal texture analysis. Optical flow analysis is used to extract motion information of texture points from adjacent frames and form a motion vector field. The main texture direction is used to characterize the main extension direction of the road surface texture. The longitudinal axis of vehicle travel is used as a reference direction for determining the texture direction. The included angle feature is used to quantify the angular relationship between the main texture direction and the longitudinal axis of vehicle travel. The longitudinal texture, transverse texture, and non-directional texture are used to represent road surface textures that are consistent with the vehicle travel direction, approximately perpendicular to the vehicle travel direction, and do not have a clear directional bias, respectively.

[0150] In practical implementation, the vehicle's forward-facing camera continuously acquires road surface images and forms a sequence of adjacent frames. The system performs optical flow calculations on the pixel grayscale changes between adjacent frames to obtain the displacement direction and magnitude of each road surface texture point in the image plane, thereby constructing a motion vector field. Subsequently, the motion vector field is statistically aggregated, and the main direction of the texture is obtained using a main direction estimation method. The angle between this main direction and the vehicle's longitudinal axis is calculated to obtain the angle feature. When the angle feature is less than a first threshold, the road surface texture direction is determined to be longitudinal texture; when the angle feature is within the range of a second threshold, it is determined to be lateral texture; and when the angle feature is discrete and there are no significant directional peaks, it is determined to be non-directional texture. The first and second thresholds can be calibrated according to the vehicle's installation posture, camera field of view, and road surface resolution. In practical applications, other threshold configurations can also be selected, which are not limited in this application.

[0151] This texture direction recognition method reflects the temporal motion characteristics of road surface texture through continuous frame optical flow and completes classification by combining the included angle features, so that the relationship between the main direction of texture and the vehicle driving direction can be stably quantified, thereby providing a road surface direction basis for subsequent lateral control parameter matching.

[0152] By adopting the above method, texture direction recognition can still be achieved based on motion vector field and angle features when the road surface texture is affected by changes in illumination, local occlusion or static texture interference. This makes the determination results of longitudinal texture, transverse texture and non-directional texture more stable, and provides reliable input for the dynamic adjustment of control parameters such as feedforward gain.

[0153] Optional, Figure 9 A flowchart illustrating the lateral control method based on road feature recognition provided in this application embodiment. Figure 8 ,like Figure 9 As shown, road surface features are extracted from the road surface image to obtain the road surface image features, including:

[0154] S901, construct gray-level co-occurrence matrices from multiple directions respectively, and calculate texture contrast features in multiple directions;

[0155] S902, calculate the mean of multiple texture contrast features to obtain the contrast parameters;

[0156] S903 maps contrast parameters to road surface roughness through pre-calibration relationships.

[0157] Among them, the gray-level co-occurrence matrix is ​​used to characterize the joint distribution relationship of gray levels in the road surface image at a specified spatial distance and direction, the texture contrast features in multiple directions are used to reflect the degree of gray level change in different directions, the contrast parameter is used as an intermediate quantity of road surface roughness, and the pre-calibration relationship is used to establish the correspondence mapping between the contrast parameter and the road surface roughness.

[0158] In practical implementation, road surface images are captured by an onboard camera and sent to an image processing unit. The image processing unit performs grayscale conversion and necessary denoising on the road surface area of ​​interest, and then constructs grayscale co-occurrence matrices in multiple directions at a preset spatial distance. These directions can correspond to longitudinal, transverse, and several diagonal directions. When calculating texture contrast features from the grayscale co-occurrence matrices in each direction, a weighted summation of the square of the grayscale difference and the co-occurrence probability is used to obtain the contrast value for the corresponding direction. The contrast values ​​from multiple directions are then averaged to form a contrast parameter that comprehensively characterizes the undulation and texture dispersion of the road surface. The pre-calibration relationship can be established using test road surface samples, which include asphalt roads, concrete roads, repaired sections, and locally uneven sections with different levels of roughness. A correspondence table or fitting function between the contrast parameter and roughness is obtained through offline calibration. The image processing unit outputs the road surface roughness value based on this correspondence table or fitting function. In practical applications, other models of the image processing unit can also be selected, and this application does not limit this.

[0159] This processing method uses texture statistics from multiple directions to characterize the surface condition of the road surface, and forms a stable contrast parameter through the mean. Then, it outputs roughness through a pre-calibrated relationship, so that the road surface roughness can be provided to the subsequent lateral control parameter matching module in a unified numerical form.

[0160] With this method, road surface roughness is determined by multi-directional texture information, the influence of single-directional noise on the results is weakened, and the roughness output has a better correspondence with the actual road surface undulation state, thus providing a more stable visual basis for parameter adjustment in lateral control.

[0161] In one possible embodiment, road surface feature extraction is performed on the road surface image to obtain road surface image features, including: performing multi-layer wavelet transform on the road surface image to obtain multiple high-frequency sub-bands; extracting the energy distribution ratio of the multiple high-frequency sub-bands; and obtaining the road surface roughness based on the energy distribution ratio.

[0162] Multi-level wavelet transform is a processing method for multi-scale frequency domain analysis of road surface images. It decomposes the image into low-frequency and high-frequency components at different decomposition levels to obtain multiple high-frequency sub-bands that characterize detailed changes. These high-frequency sub-bands carry information about abrupt texture changes, subtle undulations, and local discontinuities on the road surface, with different sub-bands corresponding to high-frequency details at different scales. The energy distribution ratio characterizes the energy proportion of each high-frequency sub-band. It is typically obtained by summing the squares of the coefficients of each high-frequency sub-band, and then normalizing the energy of each sub-band with the total energy of all high-frequency sub-bands to form the energy distribution ratio. Road surface roughness is determined based on this ratio and a preset mapping relationship. A higher energy proportion in the high-frequency sub-bands indicates more pronounced road surface details and a correspondingly higher road surface roughness.

[0163] In the specific implementation, the road surface image acquired by the vehicle-mounted camera is first processed by grayscale conversion and size normalization, and then input into the wavelet decomposition module for multi-level wavelet transform. This wavelet decomposition module can employ two-dimensional discrete wavelet transform to decompose the image layer by layer, obtaining multiple high-frequency sub-bands. The wavelet basis used can be Daubechies wavelet, Haar wavelet, or Symlets wavelet; in practical applications, other models can also be selected, and this application does not impose any limitations on this. Subsequently, the energy value of each high-frequency sub-band is calculated, and each energy value is divided by the sum of the energies of all high-frequency sub-bands to obtain the energy distribution ratio. The system further inputs this ratio into the road surface roughness mapping model. The mapping model can be established from pre-calibrated samples or implemented through a lookup table, thereby outputting the corresponding roughness value.

[0164] This method separates high-frequency details from road surface images using multi-layer wavelet transform, and then quantifies the surface undulations by utilizing the energy proportion of high-frequency subbands. This allows the roughness recognition results to directly reflect the strength of road texture and the degree of local unevenness. The resulting road roughness can be used as input for lateral control parameter matching, enabling the control parameters to adjust according to changes in road conditions, thereby improving the stability of lateral control and the consistency of trajectory tracking.

[0165] In one possible embodiment, after performing multi-source feature perception on the obtained road surface image, the method further includes: constructing a sliding time window to cache visual road surface features for the current period and several historical periods; inputting the cached visual road surface features within the sliding time window into a lightweight temporal prediction network; outputting road surface feature predictions for multiple future control periods; and initiating pre-tuning of lateral control parameters based on the road surface feature predictions.

[0166] The sliding time window is a temporal cache unit used to cache visual road surface features of the current cycle and several historical cycles. Its content is updated as the control cycle progresses, thus forming a continuous sequence of visual road surface features. The lightweight temporal prediction network is a predictive model used to extrapolate trends from the feature sequence. The model can be implemented using convolutional gating units, lightweight recurrent neural networks, or equivalent structures to output road surface feature predictions for multiple future control cycles while meeting on-board real-time requirements. The road surface feature predictions characterize possible changes in road surface adhesion, roughness, or texture direction within several future control cycles. Pre-tuning involves adjusting the lateral control parameters in advance based on these predictions, ensuring that the control parameters are updated before actual road surface changes occur.

[0167] In its implementation, after the vehicle completes road surface image acquisition and multi-source feature perception, the controller writes the visual road surface features extracted in the current cycle into a sliding time window, while simultaneously removing the earliest historical features within the window to maintain a constant window length. The window length can correspond to multiple consecutive control cycles, preserving both short-term fluctuation information and road surface change trend information. Subsequently, the feature vectors arranged chronologically within the window are input into a lightweight temporal prediction network, which outputs feature prediction values ​​for multiple future control cycles. These prediction values ​​may include the probability of road surface type changes, the trend of adhesion coefficient changes, the magnitude of roughness changes, and the texture direction offset. After the network outputs, the control system pre-corrects the proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency based on the prediction values, and sends the corrected parameters to the subsequent lateral control link. In this application, the specific number of layers, the number of hidden units, and the parameter scale of the lightweight temporal prediction network can be set according to the onboard computing power configuration. In practical applications, other network models can also be selected, and this application does not limit this.

[0168] By employing the above methods, the system can complete parameter preparation before the actual changes in road conditions, ensuring greater consistency between lateral control parameters and future road feature changes. This reduces control deviations caused by parameter lag and makes lateral tracking of the vehicle smoother on dry, wet, rough, or abruptly textured road surfaces. This approach also mitigates the impact of single-frame feature fluctuations on control parameter matching, enabling more stable lateral control responses in scenarios involving continuous road transitions.

[0169] In one implementation, the step of estimating the road adhesion coefficient based on road surface image features and vehicle kinematic parameters can be solved using an Extended Kalman Filter (EKF) state observer. Specifically, the system establishes a nonlinear tire model and a state observer, and iteratively updates the vehicle lateral stiffness and road adhesion coefficient state variables in real time within the EKF framework. Through a recursive estimation mechanism, the transient noise and true dynamic response of wheel speed and inertial measurement data are effectively separated, enabling the adhesion coefficient estimation to have faster convergence speed and higher disturbance rejection accuracy in the tire nonlinear slip region, and outputting structured adhesion state quantities for subsequent mapping modules to use.

[0170] In one implementation, the step of fusion evaluation based on multi-source evidence theory can be further enhanced by introducing a dynamic confidence scaling mechanism for sensor health and a cross-validation mechanism for prior map data. The system's built-in self-diagnostic subroutine evaluates in real time the clarity of camera images, the proportion of soiled area, and the zero-bias stability and noise variance of the IMU and wheel speed sensors, dynamically mapping the health score to the basic probability allocation scaling coefficient of the corresponding evidence body. When the fusion confidence level falls below a safety threshold, the system automatically retrieves the labeled road surface type and historical friction coefficient data for the corresponding road segment from the high-precision map, calculates the matching similarity between the prior data and the current perception result, and if the similarity meets secondary conditions, a low-rate conservative smooth transition strategy is activated to prevent the system from falling into a complete standstill and maintain controllable adjustment capabilities.

[0171] In one implementation, the step of matching lateral control parameters based on road surface characteristics can employ a continuous parameter mapping method based on B-spline surface fitting. The system utilizes discrete calibration points to construct a high-dimensional combination of B-spline basis functions, forming a continuous smooth hypersurface covering the entire vehicle speed and adhesion coefficient range. Control parameters are calculated and output by projecting real-time feature vectors onto the hypersurface. Simultaneously, a lateral stability margin criterion and actuator action rate limitation model based on a nonlinear two-degree-of-freedom vehicle model are established at the output end. The generated proportional, derivative, feedforward gains, and filter cutoff frequencies undergo real-time envelope detection. When any parameter exceeds the boundary of the tire sideslip linear zone or the steering gear mechanical limit, a priority-decreasing amplitude limiting and scaling algorithm is triggered for safety truncation, ensuring that the algorithm's optimization results are strictly constrained within the vehicle's physical domain safety boundaries.

[0172] In one implementation, the smoothing step for the lateral control parameters can employ a control loop error-coupled variable time constant smoother. The smoother's time constant dynamic response function is connected in parallel to the lateral tracking error and its first derivative as modulation input. When the tracking error is large or changes drastically, the smoothing time constant is automatically extended to force deceleration parameter updates; when the error converges rapidly, the time constant is shortened to accelerate the response. The system continuously caches historical parameter sequences during the smoothing transition. If the smoothed parameters fail to converge stably to the target neighborhood within multiple consecutive control cycles, or if the vehicle yaw rate variance continuously exceeds the stability threshold, the baseline safety zone switching logic is immediately triggered, forcibly pulling the control parameters back to the preset conservative safety zone and locking the adaptive adjustment channel until the operating condition recovers, providing a bottom-line safety guarantee for the algorithm.

[0173] In one implementation, the step of identifying the direction of road surface texture can employ a multi-scale Gabor filter bank and a phase correlation analysis algorithm. The system design uses Gabor convolution kernels covering different spatial frequencies and orientation angles to perform parallel response calculations on the preprocessed road surface image. By statistically analyzing the maximum response amplitude and phase distribution of filters in each direction, the main direction vector and spatial periodic features of the road surface texture are accurately extracted. When the texture is identified as having a specific direction and its spatial period falls within the effective response bandwidth of the lateral controller, the system synchronously injects a directional notch filter matching the texture period at the lateral error signal input. This filters out the pseudo-lateral deviation caused by periodic road surface geometric excitation from the control loop in real time, achieving active decoupling and interference suppression from the signal transmission source.

[0174] In one possible embodiment, Figure 10 A flowchart of the adaptive lateral control parameter method based on road feature recognition provided in the embodiments of this application is shown below. Figure 10 As shown, the solution includes the following process steps:

[0175] S1, Multi-source road feature perception. The vehicle acquires images of the road ahead through an onboard forward-facing camera with a sampling frequency of no less than 30Hz. At the same time, it acquires four-wheel rotation speed information through wheel speed sensors, acquires vehicle lateral acceleration and yaw rate through an inertial measurement unit (IMU), and acquires steering wheel angle and angular velocity through a steering angle sensor. The sampling frequency of the above sensors is no less than 100Hz.

[0176] S2, Online Road Surface Feature Recognition. Preprocessing of the acquired road surface images includes distortion correction, brightness equalization, and region of interest extraction. A lightweight convolutional neural network is used for semantic segmentation of the road surface images to identify road surface types, including five categories: dry asphalt, wet asphalt, cement pavement, gravel pavement, and icy / snow pavement. Simultaneously, the road adhesion coefficient is estimated based on wheel speed difference and IMU data. The calculation formula is as follows:

[0177] μest =sqrt(a y ²+(v x ·ψ̇)²) / g·K corr , where μ est a is an estimated value of the road surface adhesion coefficient. y For lateral acceleration, v x Let ψ be the longitudinal speed, ψ̇ be the yaw rate, g be the acceleration due to gravity, and K be the acceleration due to gravity. corr The correction coefficient ranges from 0.85 to 1.15. Furthermore, the road surface roughness index is calculated based on image texture analysis, using the contrast features of the Gray-Level Co-occurrence Matrix (GLCM).

[0178] Contrast=Σ(i,j)|ij|²·P(i,j), where P(i,j) is the probability value of position (i,j) in the gray-level co-occurrence matrix, and the directionality of road surface texture is identified based on optical flow analysis of continuous frame images to distinguish longitudinal texture, transverse texture and non-directional texture.

[0179] S3, Feature Fusion and Confidence Assessment. The visual recognition results and dynamic estimation results are fused using DS evidence theory. The confidence of the fusion result is calculated. When the confidence is below a threshold of 0.6, the current control parameters remain unchanged. When the confidence is above the threshold, the final road surface feature recognition result is output, including road surface type, adhesion coefficient, roughness, and texture direction.

[0180] In an optional embodiment, the processing flow for online road feature recognition is as follows: Figure 11 As shown, the system first inputs road surface images along with wheel speed / IMU data. The road surface images undergo preprocessing, including distortion correction, brightness equalization, and region of interest extraction. Then, a lightweight convolutional neural network is used for semantic segmentation to identify road surface types, including dry asphalt, wet asphalt, cement pavement, gravel pavement, and icy / snow pavement. Simultaneously, the road adhesion coefficient is estimated based on wheel speed differences and IMU data. After obtaining the road surface features, the visual recognition results and dynamic estimation results are fused using DS evidence theory. The confidence level of the fused result is calculated. When the confidence level is below a threshold (which can be set to 0.6), the current control parameters remain unchanged. When the confidence level is above the threshold, the final road surface feature recognition results are output, including: road surface type, adhesion coefficient, roughness, and texture direction.

[0181] S4, Lateral control parameter lookup and interpolation. A road feature-control parameter mapping table (equivalent to the control parameter mapping data mentioned above) is pre-established and stored in the vehicle controller's non-volatile memory. The mapping table structure is as follows: Figure 12As shown, the mapping table dimensions include 5 road surface types, adhesion coefficients from 0.3 to 1.0, and vehicle speeds from 0 to 120 km / h. The mapping table output parameters include proportional gain Kp, differential gain Kd, feedforward gain Kff, and low-pass filter cutoff frequency fc. Based on the real-time identified road surface features, multi-dimensional linear interpolation is performed in the mapping table to obtain the optimal control parameters under the current operating conditions.

[0182] S5, smooth transition of lateral control parameters. To avoid vehicle instability caused by abrupt parameter changes, a first-order low-pass filter is used to smooth the four obtained control parameters. The calculation formula is as follows:

[0183] Kout(t) = Kout(t-1) + α•(Ktable - Kout(t-1)), where Kout(t) is the output parameter at time t, Ktable is the target parameter obtained from the lookup table, and α is the smoothing coefficient, ranging from 0.05 to 0.2, dynamically adjusted according to the rate of change of road features. When road features change abruptly, the smoothing coefficient is automatically increased to 0.15 to 0.2 to accelerate parameter convergence; when road features are stable, the smoothing coefficient is decreased to 0.05 to 0.1 to reduce parameter fluctuations. The effect of smoothing the parameters is as follows: Figure 13 As shown.

[0184] S6, Lateral Controller Execution. The adjusted control parameters are input to the lateral controller, which can be a PID controller, model predictive controller, or quadratic regulator. The controller calculates the target deflection angle based on the deviation between the desired path and the actual position, executes steering control through the electric power steering system (EPS), and monitors the control effect in real time. When control divergence or vehicle instability occurs, a degradation strategy is triggered.

[0185] This embodiment improves control stability by real-time identification of the road surface adhesion coefficient and adjustment of the control gain. On slippery roads, the proportional gain can be reduced by 20% to 30%, effectively avoiding oversteering. Actual test data shows that the maximum lateral deviation decreased from 0.28m to 0.12m, an improvement of 57%. At the same time, ride comfort is improved, with the peak lateral acceleration decreasing from 2.6m / s² to 1.8m / s², an improvement of 31%. The occupant's subjective evaluation increased from 3.2 points to 4.5 points.

[0186] In an optional embodiment, the lateral control of the vehicle via the lateral controller also includes safety monitoring and degradation processing steps. The system monitors the output command amplitude of the lateral controller, the variance of the vehicle's actual yaw rate, and the quality of sensor data in real time. When lateral control divergence is detected, the yaw rate variance continuously exceeds the safety threshold, or the image command evaluation module detects camera obstruction causing the visual recognition confidence to drop to 0.3, a system degradation strategy is triggered. The degradation strategy is specifically executed as follows: the control parameter adjustment frequency is reduced from the normal operating frequency of 10Hz to the degradation frequency of 2Hz; the lateral controller switches to a dynamic estimation mode that relies solely on wheel speed sensors and inertial measurement units for parameter calculation; and simultaneously, a road surface recognition function limitation warning message is sent to the human-machine interface to remind the driver to maintain attention. When the image quality recovers and the sensor data confidence recovers, the system automatically and smoothly switches back to the normal control mode and re-enables the parameter adaptive mapping process based on visual road surface features.

[0187] In an optional embodiment, Figure 14 The following is a system hardware architecture block diagram of the adaptive lateral control parameter system based on road feature recognition according to an embodiment of this application, as shown in the figure. Figure 14 As shown, the system includes a perception layer, a processing layer, and an execution layer.

[0188] The perception layer includes hardware components such as a forward-facing camera, wheel speed sensors, an IMU, and a steering angle sensor, which are used by the system to collect signals, such as collecting road images through the forward-facing camera and collecting wheel speeds through the wheel speed sensors.

[0189] The processing layer includes a road surface recognition module and a parameter adjustment module, which are used to perform data processing actions such as image preprocessing, CNN inference, dynamics calculation, feature fusion, table lookup interpolation, and smooth transition.

[0190] The execution layer includes a lateral control module, which identifies the deviation between the desired path and the actual position through the path tracking controller. Based on this, the lateral control module calculates the target steering angle, executes steering control through the EPS actuator, and monitors the control effect in real time. When control divergence or vehicle instability occurs, a degradation strategy is triggered.

[0191] Meanwhile, the system also includes a security monitoring module that runs through the perception layer, processing layer and execution layer. The security monitoring module is used to assess the confidence of the fusion results, detect anomalies and determine whether to implement a degradation strategy.

[0192] The following detailed description of this application is based on several specific embodiments. This embodiment is a modification of a mass-produced L2+ intelligent driving system. The vehicle configuration includes a 1280*720@30fps forward-facing camera, four-wheel speed sensors, a 6-axis IMU, an EPS steering system, and a computing platform based on the Horizon J5 automotive-grade chip with a computing power of 128 T0PS.

[0193] Example 1: Driving from a dry asphalt road onto a wet and slippery road.

[0194] Scenario description: The vehicle is traveling at 80 km / h on a dry asphalt road. The road ahead is slippery due to rain, and the road adhesion coefficient drops from 0.85 to 0.55.

[0195] Processing flow: In the road perception stage, the forward-looking camera acquires one image frame every 33ms, while the IMU and wheel speed sensors acquire data every 10ms. In the road recognition stage, the convolutional neural network model performs semantic segmentation on the image, identifying that the road surface type changes from dry asphalt to wet asphalt 50m ahead. Dynamic calculations show that the estimated adhesion coefficient decreases from 0.82 to 0.58. The DS evidence theory fusion result shows the road surface type as wet asphalt with a confidence level of 0.87; and the adhesion coefficient as 0.57 with a confidence level of 0.82. In the parameter adjustment stage, the control parameters corresponding to a wet road surface adhesion coefficient of 0.6 and a vehicle speed of 80km / h are obtained by looking up a table. Kp decreases from 1.2 to 0.85 (a reduction of 29%), Kd decreases from 0.35 to 0.25 (a reduction of 29%), Kff decreases from 0.95 to 0.80 (a reduction of 16%), and fc decreases from 8Hz to 5Hz (a reduction of 37%). Due to the smoothing coefficient α for the sudden change in road features being set to 0.18, the parameters complete the transition within approximately 5 control cycles, or 50ms. The adjusted parameter input model predicts the lateral controller during the control execution phase, allowing the vehicle to smoothly pass through the road surface change area with a maximum lateral deviation of 0.12m. In contrast, the fixed parameter scheme has a maximum lateral deviation of 0.28m and exhibits slight oscillations.

[0196] Performance comparison data: Maximum lateral deviation: In this invention, the solution was 0.12m, while the fixed parameter solution was 0.28m, an improvement of 57%; Peak lateral acceleration: In this invention, the solution was 1.8m / s², while the fixed parameter solution was 2.6m / s², an improvement of 31%; Number of steering wheel corrections: In this invention, the solution was 3, while the fixed parameter solution was 8, an improvement of 62%; Passenger subjective evaluation: In this invention, the solution was 4.5 points, while the fixed parameter solution was 3.2 points, an improvement of 41%.

[0197] Example 2: Scene of driving from a cement road onto a gravel road.

[0198] Scenario description: The vehicle is traveling at 60 km / h on a cement road. The road ahead is a construction area, and the road surface has changed to a gravel road with a significantly increased roughness.

[0199] Processing flow: In the road surface recognition stage, GLCM texture analysis showed an increase in contrast feature from 0.35 to 0.72. The IMU detected an increase in high-frequency vibration components, with a 3-fold increase in energy in the 10-20Hz frequency band. The fusion result showed a road surface type of gravel with a confidence score of 0.79 and a roughness with a high confidence score of 0.85. In the parameter adjustment stage, the control parameters corresponding to a vehicle speed of 60km / h on the gravel road were obtained by referring to a table. Kp decreased from 1.5 to 1.3 (a 13% reduction), Kd decreased from 0.40 to 0.22 (a 45% reduction), and fc decreased from 10Hz to 4Hz (a 60% reduction). In the control execution stage, reducing the differential gain and filter cutoff frequency effectively filtered out vibration interference, keeping the vehicle centered in the lane without any erroneous corrections due to vibration.

[0200] Example 3: Longitudinal texture scene of bridge expansion joint.

[0201] Scenario description: A vehicle is crossing a bridge at a speed of 100 km / h. The road surface has dense longitudinal expansion joints, which may easily cause lateral control misjudgment.

[0202] Processing flow: In the road surface recognition stage, optical flow analysis identifies the road surface texture direction as longitudinal, with an angle of less than 15 degrees to the driving direction. The fusion result shows a longitudinal texture direction confidence score of 0.91. In the parameter adjustment stage, a table is consulted to obtain the feedforward gain adjustment strategy corresponding to the longitudinal texture road surface, and Kff is reduced from 1.0 to 0.7, a reduction of 30%. In the control execution stage, the vehicle smoothly passes through the expansion joint area without any jerking or unevenness.

[0203] Example 4: System degradation strategy.

[0204] Scenario description: The camera was blocked by mud, causing the visual recognition function to fail.

[0205] Processing flow: In the anomaly detection phase, the image quality assessment module detects that the image blur exceeds the threshold, and the confidence level of the visual recognition result drops to 0.3. In the degradation strategy phase, the system switches to pure dynamic estimation mode, relying solely on wheel speed and IMU data. The control parameter adjustment frequency is reduced from 10Hz to 2Hz, and the HMI (Human Machine Interface) prompts the driver that the road recognition function is limited and requests continued attention. In the function recovery phase, once the image quality is restored, the system automatically switches back to normal mode, with a smooth parameter transition to avoid abrupt changes.

[0206] The following is an example of a parameter mapping table design:

[0207] When the road surface type is dry asphalt with an adhesion coefficient of 0.8 to 1.0: Kp is 1.20, Kd is 0.35, Kff is 0.95, and fc is 8Hz;

[0208] When the road surface type is dry asphalt with an adhesion coefficient of 0.6 to 0.8: Kp is 1.35, Kd is 0.38, Kff is 0.95, and fc is 8Hz;

[0209] When the road surface type is wet asphalt with an adhesion coefficient of 0.5 to 0.6: Kp is 0.85, Kd is 0.25, Kff is 0.80, and fc is 5Hz;

[0210] When the road surface type is wet asphalt with an adhesion coefficient of 0.3 to 0.5: Kp is 0.65, Kd is 0.18, Kff is 0.65, and fc is 4Hz;

[0211] When the road surface type is cement pavement and the adhesion coefficient is 0.7 to 0.9: Kp is 1.40, Kd is 0.40, Kff is 0.95, and fc is 10Hz;

[0212] When the road surface type is gravel road with an adhesion coefficient of 0.4 to 0.6: Kp is 1.30, Kd is 0.22, Kff is 0.75, and fc is 4Hz;

[0213] When the road surface type is ice and snow and the adhesion coefficient is 0.2 to 0.4: Kp is 0.45, Kd is 0.12, Kff is 0.50, and fc is 3Hz.

[0214] The following is an example of a convolutional neural network model design:

[0215] The network architecture uses MobileNetV3-Small pre-trained weights for the backbone, FPN feature pyramid for the neck, and the head for the first 5 semantic segmentation road surface classes plus background. The input resolution is 320×240, and the output resolution is 80×60. Training data comes from real-vehicle data collection and publicly available datasets, totaling approximately 50,000 frames. Data augmentation includes brightness adjustment, contrast adjustment, Gaussian noise, and motion blur. The training configuration uses PyTorch 1.13, with CrossEntropy and DiceLoss loss functions, AdamW optimizer with a learning rate of 1e-4, a batch size of 32, and 100 training epochs. The model performance is mIoU of 87.3%, inference time of 12ms, and model size of 4.2MB.

[0216] An optional software architecture in this application embodiment adopts the AUTOSARAP standard. The application layer includes road surface recognition SWC, parameter adjustment SWC, lateral control SWC, and monitoring SWC. The base layer includes camera driver, IMU driver, CAN driver, and diagnostic services. The functional safety ASIL level is ASIL-B, and the safety mechanisms include dual-core lockstep, ECC memory protection, and watchdog monitoring. The fault detection time is less than 100ms, and the fault response time is less than 50ms.

[0217] The test scenario library in this application includes 15 scenarios for dry roads, 12 scenarios for wet and slippery roads, 8 scenarios for icy and snowy roads, 10 scenarios for rough roads, and 20 scenarios for mixed roads, totaling 65 scenarios and a test mileage of 1850km. The test results show that all 65 scenarios passed the acceptance test, the lateral control accuracy compliance rate was 98.5%, the subjective evaluation of ride comfort was 4.3 points, and the system availability was 97.8%.

[0218] Figure 15 This is a schematic diagram of the lateral control device based on road feature recognition provided in an embodiment of this application. The device in this embodiment can be in the form of software and / or hardware. Figure 15 As shown in the embodiment of this application, the lateral control device 1500 based on road surface feature recognition includes: a feature sensing module 1501, a matching module 1502, and a lateral control module 1503.

[0219] The feature perception module 1501 is used to acquire road surface images in real time, perform multi-source feature perception on the obtained road surface images, and obtain visual road surface features, including: road surface type, road surface adhesion coefficient estimate, road surface roughness, and road surface texture direction.

[0220] Matching module 1502 is used to match the corresponding lateral control parameters in the control parameter mapping data based on visual road surface features;

[0221] The lateral control module 1503 is used to input lateral control parameters into the lateral controller, and the lateral controller performs lateral control on the vehicle.

[0222] This device acquires road surface images in real time and extracts visual road surface features such as road surface type, adhesion coefficient estimate, roughness, and texture direction. It can establish a correspondence between lateral control parameters and the current physical state of the road surface, and then match more suitable control parameters in a timely manner under road surface conditions such as wet, rough, or with significant texture changes. By inputting the matched lateral control parameters into the lateral controller, it can reduce gain mismatch and control oscillation caused by fixed or lagging parameters, thereby improving the stability of vehicle lateral control, trajectory tracking accuracy, and driving smoothness.

[0223] In one possible implementation, the feature sensing module 1501 is further configured to:

[0224] Road surface features are extracted from the road surface image to obtain road surface image features, which include: road surface type, road surface roughness, and road surface texture direction;

[0225] The road surface adhesion coefficient is estimated based on the road surface image features, and the estimated value of the road surface adhesion coefficient is obtained.

[0226] In one possible implementation, the feature sensing module 1501 is further configured to:

[0227] The confidence level of visual road surface features is evaluated based on the moving road surface features, which are calculated based on vehicle kinematic parameters.

[0228] When the confidence level is greater than the first threshold, the lateral control parameters are matched in the control parameter mapping data based on the visual road surface features.

[0229] In one possible implementation, the matching module 1502 is further configured to:

[0230] The index key is determined based on visual road surface features and the vehicle's real-time speed.

[0231] Multidimensional linear interpolation is performed on the control parameter mapping data based on the index key to match the lateral control parameters, which include: proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency.

[0232] In one possible implementation, the matching module 1502 is further configured to:

[0233] The feedforward gain is dynamically adjusted based on the direction of the road surface texture.

[0234] In one possible implementation, the lateral control module 1503 is further configured to:

[0235] The transverse control parameters are smoothed by a first-order low-pass filter to obtain the processed transverse control parameters.

[0236] The processed lateral control parameters are input into the lateral controller, which then performs lateral control on the vehicle.

[0237] In one possible implementation, the lateral control module 1503 is further configured to:

[0238] Real-time calculation of the feature change rate of visual road surface features;

[0239] The smoothing coefficient is dynamically adjusted based on the characteristic rate of change.

[0240] The transverse control parameters are smoothed using a first-order low-pass filter based on the adjusted smoothing coefficient.

[0241] In one possible implementation, the feature sensing module 1501 is further configured to:

[0242] Optical flow analysis is performed on consecutive frames of road surface images to extract the motion vector field of road surface texture points;

[0243] Calculate the angle between the main direction of the texture and the longitudinal axis of the vehicle's movement;

[0244] The direction of road surface texture is identified based on the included angle feature, where the direction of road surface texture includes: longitudinal texture, transverse texture and non-directional texture.

[0245] In one possible implementation, the feature sensing module 1501 is further configured to:

[0246] Gray-level co-occurrence matrices are constructed from multiple directions, and texture contrast features in multiple directions are calculated.

[0247] Calculate the mean of multiple texture contrast features to obtain the contrast parameters;

[0248] The contrast parameter is mapped to road surface roughness through a pre-calibration relationship.

[0249] The lateral control device based on road feature recognition provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0250] Figure 16 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 16 As shown, the electronic device 1600 provided in this embodiment includes at least one processor 1601 and a memory 1602. Optionally, the device 1600 further includes a communication component 1603. The processor 1601, memory 1602, and communication component 1603 are connected via a bus.

[0251] In a specific implementation, at least one processor 1601 executes computer execution instructions stored in memory 1602, causing at least one processor 1601 to perform the above-described method.

[0252] The specific implementation process of processor 1601 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0253] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0254] The memory may include high-speed memory (Random Access Memory, RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0255] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0256] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0257] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0258] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0259] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in application-specific integrated circuits (ASICs). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0260] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0261] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0262] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0263] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0264] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0265] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A lateral control method based on road surface feature recognition, characterized in that, include: Real-time acquisition of road surface images, multi-source feature perception of the obtained road surface images to obtain visual road surface features, wherein the visual road surface features include: road surface type, road surface adhesion coefficient estimate, road surface roughness, and road surface texture direction. Match the corresponding lateral control parameters in the control parameter mapping data based on the visual road surface features; The lateral control parameters are input into the lateral controller, which then performs lateral control on the vehicle.

2. The method according to claim 1, characterized in that, The process of performing multi-source feature perception on the obtained road surface image to obtain visual road surface features includes: The road surface image is subjected to road surface feature extraction to obtain road surface image features, wherein the road surface image features include: the road surface type, the road surface roughness, and the road surface texture direction; The road surface adhesion coefficient is estimated based on the road surface image features to obtain the estimated value of the road surface adhesion coefficient.

3. The method according to claim 1, characterized in that, After performing multi-source feature perception on the obtained road surface image to obtain visual road surface features, the method further includes: The confidence level of the visual road surface features is evaluated based on the moving road surface features, wherein the moving road surface features are calculated based on vehicle kinematic parameters; If the confidence level is greater than a first threshold, the lateral control parameter is matched in the control parameter mapping data according to the visual road surface features.

4. The method according to claim 2, characterized in that, The step of matching the lateral control parameters in the control parameter mapping data based on the visual road surface features includes: The index key is determined based on the visual road surface features and the vehicle's real-time speed; Multidimensional linear interpolation is performed on the control parameter mapping data based on the index key to match the lateral control parameters, wherein the lateral control parameters include: proportional gain, derivative gain, feedforward gain, and low-pass filter cutoff frequency.

5. The method according to claim 4, characterized in that, After matching the lateral control parameters in the control parameter mapping data based on the visual road surface features, the method further includes: The feedforward gain is dynamically adjusted based on the direction of the road surface texture.

6. The method according to claim 4, characterized in that, The step of inputting the lateral control parameters into the lateral controller and performing lateral control of the vehicle through the lateral controller includes: The lateral control parameters are smoothed by a first-order low-pass filter to obtain the processed lateral control parameters. The processed lateral control parameters are input into the lateral controller, which then performs lateral control on the vehicle.

7. The method according to claim 6, characterized in that, The lateral control parameters are smoothed using a first-order low-pass filter, including: Real-time calculation of the feature change rate of the visual road surface features; The smoothing coefficient is dynamically adjusted based on the rate of change of the described features; The lateral control parameters are smoothed using the first-order low-pass filter based on the adjusted smoothing coefficient.

8. The method according to claim 2, characterized in that, The road surface image is subjected to road surface feature extraction to obtain road surface image features, including: Optical flow analysis is performed on consecutive frames of the road surface image to extract the motion vector field of the road surface texture points; Calculate the angle between the main direction of the texture and the longitudinal axis of the vehicle's movement; The direction of the road surface texture is identified based on the included angle feature, wherein the direction of the road surface texture includes: longitudinal texture, transverse texture and non-directional texture.

9. The method according to claim 2, characterized in that, The road surface image is subjected to road surface feature extraction to obtain road surface image features, including: Gray-level co-occurrence matrices are constructed from multiple directions, and texture contrast features in the multiple directions are calculated. The mean of multiple texture contrast features is calculated to obtain the contrast parameter; The contrast parameter is mapped to the road surface roughness through a pre-calibration relationship.

10. A lateral control device based on road surface feature recognition, characterized in that, include: The feature perception module is used to acquire road surface images in real time, perform multi-source feature perception on the obtained road surface images, and obtain visual road surface features, wherein the visual road surface features include: road surface type, road surface adhesion coefficient estimate, road surface roughness, and road surface texture direction. The matching module is used to match the corresponding lateral control parameters in the control parameter mapping data based on the visual road surface features; The lateral control module is used to input the lateral control parameters into the lateral controller, and to perform lateral control of the vehicle through the lateral controller.