A robust control method and system for human-like autonomous driving based on visual reference points
By using a visual reference point-based approach, image data is acquired through a forward-looking camera to identify target vehicles and lane lines, and longitudinal and lateral control feature data is constructed. This solves the problems of high cost and instability in existing autonomous driving technologies, and achieves low-cost and highly robust autonomous driving control.
Patent Information
- Application Number
- CN202610794301.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-03
- Publication Date
- 2026-11-17
AI Technical Summary
Existing autonomous driving technologies rely on high-cost sensors and complex computing platforms, and their detection is unstable in complex environments, making them difficult to popularize. The black-box model lacks generalization ability and interpretability.
By using a visual reference point-based approach, image data is acquired using a forward-looking camera to identify target vehicles, lane lines, and road areas. This constructs longitudinal and lateral control feature data, which, combined with dynamic compensation and fuzzy control models, generates target acceleration and steering angle to achieve vehicle cooperative control.
It improves the stability and robustness of vehicles in complex environments, reduces sensor-dependent costs, and enhances the interpretability and generalization of control.
Smart Images

Figure CN123104684A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of autonomous driving technology, specifically relating to a highly robust control method and system for human-like autonomous driving based on visual reference points. Background Technology
[0002] With the development of intelligent driving technology, autonomous driving systems are gradually being applied to passenger vehicles, logistics transportation, and unmanned delivery. Current autonomous driving technologies typically employ a perception scheme combining LiDAR, millimeter-wave radar, high-precision maps, and multi-camera fusion, along with deep neural networks for environmental understanding and driving decisions, to achieve functions such as longitudinal following, lateral lane keeping, and obstacle avoidance in complex scenarios. However, existing technologies still have significant limitations in practical applications. On the one hand, current solutions rely heavily on high-precision sensors and complex computing platforms, requiring expensive LiDAR, GPU computing units, and high-precision map systems, resulting in high overall hardware costs and hindering large-scale adoption in ordinary vehicles. On the other hand, multi-sensor fusion systems are sensitive to sensor installation accuracy, calibration errors, and changes in the external environment. In complex scenarios such as rain, snow, strong light, tunnels, and obstructions, they are prone to detection jitter, misidentification, and control instability, thus affecting vehicle driving safety and system robustness.
[0003] Furthermore, some mainstream autonomous driving technologies employ end-to-end deep learning models to directly implement driving decisions from image input to control output, such as black-box models trained on large-scale driving data. While these approaches can achieve high driving fit in specific scenarios, their heavy reliance on massive training data and specific driver behavior samples leads to poor model interpretability, debugging difficulties, and insufficient generalization ability. When road conditions, vehicle posture, or lighting conditions change, black-box models are prone to unpredictable control outcomes. Human drivers, on the other hand, typically do not rely on complex sensors or high-precision maps. Instead, they observe a fixed visual reference area in front of the vehicle and make intuitive judgments based on road boundaries, lane lines, and the relative position of the vehicle in front, thus achieving stable driving. Therefore, how to construct a human-like autonomous driving control method and system based on fixed visual reference points, possessing high robustness, low cost, interpretability, and without relying on massive data training has become a crucial technical problem urgently needing to be solved in the current autonomous driving technology field. Summary of the Invention
[0004] To achieve the above-mentioned objectives, this invention provides the following technical solution: a human-like autonomous driving highly robust control method based on visual reference points, comprising the following steps: Step S1: Acquire road image data in front of the vehicle through the front view image acquisition unit, and obtain longitudinal reference points and lateral reference points based on the fixed visual reference point calibration mechanism. At the same time, identify and process the target vehicle, lane lines and road area in front to generate target bottom edge coordinate data, lane center coordinate data and road drivable area data. Step S2: Construct a longitudinal proximity parameter based on the target bottom edge coordinate data and the longitudinal reference point, and construct a lateral deviation parameter based on the lane center coordinate data and the lateral reference point. At the same time, combine the vehicle operating status data to perform dynamic compensation processing on the longitudinal and lateral deviations, and generate compensated longitudinal and lateral control feature data. Step S3: Based on the compensated longitudinal and lateral control feature data, construct a longitudinal state decision model and a lateral fuzzy control model, and combine the curve curvature parameters, vehicle speed parameters and historical trajectory projection data to generate the target acceleration and target steering wheel angle; Step S4: Input the target acceleration and the target steering wheel angle into the vehicle actuator, and construct a fused drivable area confidence map based on historical trajectory mask data and semantic segmentation data to perform coordinated control of the vehicle's longitudinal driving and lateral steering processes, thereby completing the highly robust control of human-like autonomous driving.
[0005] Preferably, step S1 further includes: The road image frames are captured in real time by a forward-looking camera, and the target detection processing is performed on the road image frames to obtain the bottom vertical coordinate of the vehicle detection box in front. Lane edge features are extracted using a lane line detection network, and the lane center coordinates are obtained by fitting the left and right lane boundaries. A fixed marker is placed at the leading edge of the vehicle hood while the vehicle is stationary, and a longitudinal reference point is determined based on the longitudinal pixel position of the fixed marker in the image. Define the center position of the image as the horizontal reference point; A semantic segmentation network is used to perform pixel-level classification of road areas, generating a confidence map of drivable road areas.
[0006] Preferably, step S2 further includes: Based on the longitudinal coordinates of the target vehicle's bottom edge and the longitudinal reference point, construct the longitudinal proximity parameters and calculate the longitudinal rate of change; The lateral deviation parameters are constructed based on the lane center coordinates and the lateral reference point, and the lateral change rate is calculated. Common-mode suppression processing is performed on the longitudinal drift of the image based on the change in vehicle pitch angle, and a compensated longitudinal proximity parameter is constructed. The lane line coordinates are rotated and compensated based on the vehicle roll angle, and lateral compensation deviation parameters are constructed.
[0007] Preferably, step S3 further includes: Based on the vehicle's current speed Constructing a dynamic longitudinal threshold: ; ; ; in, , , These represent the basic threshold parameters, , Indicates the speed adjustment coefficient; Based on longitudinal proximity parameters Constructing vertical semantic states: when When this occurs, it is determined to be a long-distance state; when At that time, it is determined to be a mid-range state; when At that time, it is determined to be a close-range situation; when When this occurs, it is determined to be a dangerous approach situation; The longitudinal proximity parameter is smoothed using a time-filtering model: ; in, Indicates the filter weight coefficients; The braking acceleration is constructed based on the smoothed longitudinal proximity parameters and the rate of change: ; in, Indicates the maximum braking acceleration. , Indicates the control gain coefficient; Constructing road curve curvature parameters using the rate of change of lane line curvature: ; in, Indicates the number of sampling points; Based on the curvature parameters Curve correction for longitudinal threshold: ; According to the vehicle pitch angle Correcting the longitudinal reference point: ; in, Indicates the pitch compensation coefficient; According to the curve rating Cooperative constraints on vehicle target speed: ; ; in, Indicates the base speed. Indicates the curve limit coefficient. Indicates the set cruise speed; According to the braking acceleration Target speed and the vehicle's current speed Construct target longitudinal control acceleration : ; in, Indicates the speed tracking gain coefficient; Based on the lateral compensation deviation parameter Horizontal deviation change rate and cornering level Construct the target steering wheel angle: ; in, This represents the lateral deviation control coefficient. This represents the horizontal dynamic compensation coefficient. This represents the cornering compensation coefficient.
[0008] Preferably, step S4 further includes: The target acceleration generated in step S3 Steering wheel angle relative to the target The signals are output to the vehicle actuators and converted into throttle control signals, brake control signals, and steering control signals, respectively. Record vehicle historical trajectory points And by using extrinsic and intrinsic parameters from the camera, the historical trajectory points are projected onto the current image coordinate system: ; in, This represents the camera intrinsic parameter matrix. This represents the coordinate transformation matrix from the vehicle body to the camera; Construct a historical passable mask based on the projected historical trajectory area: ; And construct trajectory confidence weights using a time decay function: ; in, Indicates the initial weight parameters. Represents the time decay constant; Obtain the confidence map of the drivable region generated by semantic segmentation. The drivable area confidence map is then fused with the historical drivable mask to generate a fused drivable area confidence map. ; According to the curve rating Coordinated restrictions on the maximum permissible speed of vehicles: ; And according to the vehicle's set cruise speed Generate target vehicle speed: ; in, Indicates the basic vehicle speed parameter. Indicates the curve constraint coefficient; Based on the fused drivable area confidence map The confidence level of the vehicle’s current driving area is verified. When the confidence level of the corresponding area in the fused drivable area confidence map is lower than a preset threshold, deceleration control or braking control is triggered to improve the vehicle’s driving stability and robustness in scenarios with sudden changes in lighting, occlusion, and blurred road boundaries.
[0009] This invention also provides a robust human-like autonomous driving control system based on visual reference points, comprising the following modules: Image acquisition module, used to acquire image data of the road in front of the vehicle; The fixed reference point calibration module is used to acquire and store longitudinal and lateral reference points; The target detection module is used to detect target vehicles and output the target's bottom edge coordinate data; The lane line detection module is used to detect lane lines and output lane center coordinate data; The longitudinal and lateral deviation calculation module is used to calculate the longitudinal proximity parameter, the lateral deviation parameter, and their rate of change. The scene compensation module is used to compensate and correct longitudinal and lateral deviations based on the vehicle's pitch and roll angles. The longitudinal and lateral decision control module is used to generate target acceleration and target steering wheel angle based on dynamic thresholds, fuzzy rules, and finite state machines; The historical trajectory fusion module is used to construct a historical trajectory mask and fuse it with the semantic segmentation results; The collaborative control module is used to jointly constrain and control the vehicle's longitudinal speed and lateral steering according to the curve level. The execution module is used to output throttle, brake, and steering control signals.
[0010] The longitudinal and lateral deviation calculation module includes: The longitudinal proximity calculation unit is used to calculate the longitudinal proximity of the target vehicle according to the following formula; The lateral deviation calculation unit is used to calculate the lane center lateral deviation according to the following formula; The rate of change analysis unit is used to calculate the rate of change of longitudinal and transverse deviations and generate dynamic control trend parameters.
[0011] The historical trajectory fusion module includes: The trajectory recording unit is used to record the vehicle's historical trajectory points; The trajectory projection unit is used to map historical trajectories to the current image coordinate system using camera intrinsic and extrinsic parameters. The region fusion unit is used to generate a fused drivable region confidence map according to the following formula to improve the path robustness of vehicles in complex scenarios.
[0012] The collaborative control module includes: The curve recognition unit is used to identify the curve level based on road curvature parameters; The speed coordination control unit is used to generate the maximum permissible vehicle speed according to the following formula; The steering coordination control unit is used to adjust the steering parameters based on lateral deviation parameters, lateral change rate, and cornering grade. Output the target steering wheel angle.
[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: Enhancing the stability and anti-shake capability of visual perception benchmarks: By setting fixed visual reference points and combining pitch and roll angle compensation mechanisms, image drift caused by camera attitude changes during vehicle operation is effectively suppressed, reducing longitudinal and lateral deviation calculation errors from the source. This makes the spatial representation of target vehicles, lane lines, and drivable areas more stable and reliable, thereby improving the consistency and accuracy of overall perception data.
[0014] Enhance the dynamic robustness of longitudinal and lateral coupling control: By constructing a dual-channel control system based on proximity and deviation parameters and introducing dynamic factors such as speed, curvature and rate of change, the coordinated adjustment of longitudinal braking / acceleration and lateral steering is achieved; at the same time, by combining fuzzy state division and dynamic threshold mechanism, the vehicle can automatically adjust the control strategy according to different distance states and road curvature, thereby significantly improving the control stability and response robustness in complex traffic scenarios.
[0015] Improve drivable area identification and safety redundancy in complex environments: Construct a fused drivable area confidence map by weighted fusion of semantic segmentation results and historical trajectory projection masks, and introduce a time decay weight mechanism to enable historical driving experience to participate in current decision-making. At the same time, enhance the fault tolerance of abnormal scenarios such as sudden changes in lighting, occlusion and blurred lane boundaries, and improve the safety and continuity of path planning while reducing the risk of false detection. Attached Figure Description
[0016] Figure 1 A flowchart illustrating the method steps provided in this application; Figure 2 A schematic diagram of the system modules provided in this application. Detailed Implementation
[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0018] refer to Figure 1 This invention provides a highly robust human-like autonomous driving control method based on visual reference points, comprising the following steps: Step S1: Acquire road image data in front of the vehicle through the forward-view image acquisition unit, and obtain longitudinal reference points based on the fixed visual reference point calibration mechanism. and lateral reference points Simultaneously, it identifies and processes target vehicles, lane lines, and road areas ahead, generating target bottom edge coordinate data. Lane center coordinate data and data on drivable road areas ; Step S2: Based on the target bottom edge coordinate data With the longitudinal reference point Constructing longitudinal proximity parameters And based on the lane center coordinate data With the lateral reference point Constructing lateral deviation parameters Simultaneously, the longitudinal and lateral deviations are dynamically compensated by combining vehicle operating status data to generate compensated longitudinal and lateral control characteristic data. Step S3: Construct a longitudinal state decision model and a lateral fuzzy control model based on the compensated longitudinal and lateral control feature data, and combine them with the curve curvature parameters. Vehicle speed parameters And the target acceleration generated from historical trajectory projection data Steering wheel angle relative to the target ; Step S4: Measure the target acceleration With the target steering wheel angle By inputting the vehicle actuators and constructing a fused drivable area confidence map based on historical trajectory mask data and semantic segmentation data, the vehicle's longitudinal driving and lateral steering processes are coordinated and controlled to achieve highly robust human-like autonomous driving control.
[0019] In step S1, road image frames are acquired in real time using a forward-looking camera. The road image frame is then processed for target detection to obtain the ordinate of the bottom edge of the detection box for the vehicle ahead. ; Lane edge features are extracted using a lane line detection network, and the lane center coordinates are obtained by fitting the left and right lane boundaries. ; A fixed marker is placed at the leading edge of the vehicle hood while the vehicle is stationary, and a vertical reference point is determined based on the vertical pixel position of the fixed marker in the image: ; in, This represents the vertical coordinate of the image corresponding to the fixed marker; Define the image center position as the horizontal reference point: ; in, Indicates the image width; A semantic segmentation network is used to perform pixel-level classification of road regions, generating a confidence map of drivable road areas. ; in, Represents pixels The probability value that the area belongs to a drivable zone.
[0020] In step S2, longitudinal proximity parameters are constructed based on the longitudinal coordinates of the bottom edge of the target vehicle and the longitudinal reference point: ; And calculate the longitudinal rate of change: ; The lateral deviation parameters are constructed based on the lane center coordinates and the lateral reference point: ; And calculate the horizontal rate of change: ; Based on the change in vehicle pitch angle Common-mode suppression processing is applied to the vertical drift of the image: ; And construct the compensated longitudinal proximity parameters: ; in, This indicates the camera's focal length parameter; According to the vehicle roll angle Perform rotation compensation on the lane line coordinates: ; And construct the lateral compensation deviation parameters: ; in, This represents the tilt rotation matrix.
[0021] In step S3, based on the vehicle's current speed Constructing a dynamic longitudinal threshold: ; ; ; in, , , These represent the basic threshold parameters, , Indicates the speed adjustment coefficient; Based on longitudinal proximity parameters Constructing vertical semantic states: when When this occurs, it is determined to be a long-distance state; when At that time, it is determined to be a mid-range state; when At that time, it is determined to be a close-range situation; when When this occurs, it is determined to be a dangerous approach situation; The longitudinal proximity parameter is smoothed using a time-filtering model: ; in, Indicates the filter weight coefficients; The braking acceleration is constructed based on the smoothed longitudinal proximity parameters and the rate of change: ; in, Indicates the maximum braking acceleration. , Indicates the control gain coefficient; Constructing road curve curvature parameters using the rate of change of lane line curvature: ; in, Indicates the number of sampling points; Based on the curvature parameters Curve correction for longitudinal threshold: ; According to the vehicle pitch angle Correcting the longitudinal reference point: ; in, Indicates the pitch compensation coefficient; According to the curve rating Cooperative constraints on vehicle target speed: ; ; in, Indicates the base speed. Indicates the curve limit coefficient. Indicates the set cruise speed; According to the braking acceleration Target speed and the vehicle's current speed Construct target longitudinal control acceleration : ; in, Indicates the speed tracking gain coefficient; Based on the lateral compensation deviation parameter Horizontal deviation change rate and cornering level Construct the target steering wheel angle: ; in, This represents the lateral deviation control coefficient. This represents the horizontal dynamic compensation coefficient. This represents the cornering compensation coefficient.
[0022] In step S4, the target acceleration generated in step S3 is... Steering wheel angle relative to the target The signals are output to the vehicle actuators and converted into throttle control signals, brake control signals, and steering control signals, respectively. Record vehicle historical trajectory points And by using extrinsic and intrinsic parameters from the camera, the historical trajectory points are projected onto the current image coordinate system: ; in, This represents the camera intrinsic parameter matrix. This represents the coordinate transformation matrix from the vehicle body to the camera; Construct a historical passable mask based on the projected historical trajectory area: ; And construct trajectory confidence weights using a time decay function: ; in, Indicates the initial weight parameters. Represents the time decay constant; Obtain the confidence map of the drivable region generated by semantic segmentation. The drivable area confidence map is then fused with the historical drivable mask to generate a fused drivable area confidence map. ; According to the curve rating Coordinated restrictions on the maximum permissible speed of vehicles: ; And according to the vehicle's set cruise speed Generate target vehicle speed: ; in, Indicates the basic vehicle speed parameter. Indicates the curve constraint coefficient; Based on the fused drivable area confidence map The confidence level of the vehicle’s current driving area is verified. When the confidence level of the corresponding area in the fused drivable area confidence map is lower than a preset threshold, deceleration control or braking control is triggered to improve the vehicle’s driving stability and robustness in scenarios with sudden changes in lighting, occlusion, and blurred road boundaries.
[0023] refer to Figure 2 This invention provides a robust human-like autonomous driving control system based on visual reference points, comprising the following modules: Image acquisition module, used to acquire image data of the road in front of the vehicle; The fixed reference point calibration module is used to acquire and store longitudinal reference points. Horizontal reference point ; The target detection module is used to detect target vehicles and output the target's bottom edge coordinate data. ; The lane line detection module is used to detect lane lines and output lane center coordinate data. ; The longitudinal and lateral deviation calculation module is used to calculate the longitudinal proximity parameter. Lateral deviation parameters and its rate of change; The scene compensation module is used to adjust the vehicle pitch angle. with roll angle Compensate and correct for longitudinal and lateral deviations; The longitudinal and transverse decision control modules are used to generate target acceleration based on dynamic thresholds, fuzzy rules, and finite state machines. Steering wheel angle relative to the target ; The historical trajectory fusion module is used to construct a historical trajectory mask and fuse it with the semantic segmentation results; The collaborative control module is used to jointly constrain and control the vehicle's longitudinal speed and lateral steering according to the curve level. The execution module is used to output throttle, brake, and steering control signals.
[0024] The longitudinal and lateral deviation calculation module includes: The longitudinal proximity calculation unit is used to calculate the longitudinal proximity of the target vehicle according to the following formula: ; The lateral deviation calculation unit is used to calculate the lane center lateral deviation according to the following formula: ; The rate of change analysis unit is used to calculate the rate of change of longitudinal and transverse deviations and generate dynamic control trend parameters.
[0025] The historical trajectory fusion module includes: The trajectory recording unit is used to record the vehicle's historical trajectory points; The trajectory projection unit is used to map historical trajectories to the current image coordinate system using camera intrinsic and extrinsic parameters. The region fusion unit is used to generate a fused drivable region confidence map according to the following formula to improve the path robustness of vehicles in complex scenarios: .
[0026] The collaborative control module includes: Curve recognition unit, used to identify curves based on road curvature parameters. Identify curve grades; The speed coordination control unit is used to generate the maximum permissible vehicle speed according to the following formula: ; Steering coordination control unit, used to adjust lateral deviation parameters Horizontal variation rate and cornering level Output target steering wheel angle .
[0027] Example 3: This embodiment uses a comprehensive urban road driving scenario as an example to verify the overall robustness control method for human-like autonomous driving based on visual reference points. The vehicle is equipped with a forward-facing camera (1920×1080 resolution, 60° field of view), and visual reference points are determined under static calibration conditions, where the longitudinal reference point is... Pixels, with the horizontal reference point as Pixel.
[0028] In urban following scenarios, the system obtains the ordinate of the bottom edge of the detection box of the vehicle ahead through a target detection network. And calculate the longitudinal proximity parameter: ; When the vehicle speed is The dynamic threshold is constructed as follows: ; ; ; At this time, the detection results are obtained. The system determined it to be a mid-range state, but considering the longitudinal rate of change... ; This indicates a continuous approach trend, thus triggering a predictive control strategy to switch from the "normal" state to the "pre-deceleration" state, outputting longitudinal control acceleration to achieve smooth deceleration; ; In terms of road environment understanding, the semantic segmentation module outputs a confidence map of drivable areas. Meanwhile, the historical trajectory projection module generates a historical passable mask. And through the time decay function Construct a fused drivable area confidence map: ; Since the current semantic segmentation confidence is 0.6, while the historical trajectory weight is relatively high ( The final fusion region remains stable above 0.8, thus avoiding false braking.
[0029] In a slope environment, the IMU detects the vehicle's pitch angle. Through compensation coefficient Correct the longitudinal reference point: ; Recalculated after correction: ; Compared to uncompensated This significantly improves the accuracy of distance estimation, thereby avoiding excessive braking caused by misjudging that the distance to the vehicle in front is too close.
[0030] In terms of lateral control, the vehicle experiences lateral deviation due to crosswinds or slight road deviations. The system calculates using the lane center coordinates and a reference point: ; And combined with the rate of change With cornering level Generate the target steering wheel angle: ; This allows the vehicle to gradually return to the center of the lane.
[0031] Based on the above multi-source information fusion control process, the system can maintain stable longitudinal and lateral control outputs in complex scenarios such as urban following, slope correction, and slight deviation, achieving a balance between deceleration smoothness, trajectory consistency, and path safety, thereby improving the overall driving robustness in complex urban environments.
[0032] It should be noted that, unless otherwise specified, the embodiments and features and technical solutions in the present invention can be combined with each other.
[0033] Obviously, the embodiments described above are merely some embodiments of the present invention, not all embodiments. The accompanying drawings show preferred embodiments of the present invention, but do not limit the patent scope of the present invention. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the patent protection scope of this invention.
Claims
1. A robust control method for human-like autonomous driving based on visual reference points, characterized in that, Includes the following steps: Step S1: Acquire road image data in front of the vehicle through the front view image acquisition unit, and obtain longitudinal reference points and lateral reference points based on the fixed visual reference point calibration mechanism. At the same time, identify and process the target vehicle, lane lines and road area in front to generate target bottom edge coordinate data, lane center coordinate data and road drivable area data. Step S2: Construct a longitudinal proximity parameter based on the target bottom edge coordinate data and the longitudinal reference point, and construct a lateral deviation parameter based on the lane center coordinate data and the lateral reference point. At the same time, combine the vehicle operating status data to perform dynamic compensation processing on the longitudinal and lateral deviations, and generate compensated longitudinal and lateral control feature data. Step S3: Based on the compensated longitudinal and lateral control feature data, construct a longitudinal state decision model and a lateral fuzzy control model, and combine the curve curvature parameters, vehicle speed parameters and historical trajectory projection data to generate the target acceleration and target steering wheel angle; Step S4: Input the target acceleration and the target steering wheel angle into the vehicle actuator, and construct a fused drivable area confidence map based on historical trajectory mask data and semantic segmentation data to perform coordinated control of the vehicle's longitudinal driving and lateral steering processes, thereby completing the highly robust control of human-like autonomous driving.
2. The human-like autonomous driving high-robustness control method based on visual reference points according to claim 1, characterized in that, Step S1 further includes: The road image frames are captured in real time by a forward-looking camera, and the target detection processing is performed on the road image frames to obtain the bottom vertical coordinate of the vehicle detection box in front. Lane edge features are extracted using a lane line detection network, and the lane center coordinates are obtained by fitting the left and right lane boundaries. A fixed marker is placed at the leading edge of the vehicle hood while the vehicle is stationary, and a longitudinal reference point is determined based on the longitudinal pixel position of the fixed marker in the image. Define the center position of the image as the horizontal reference point; A semantic segmentation network is used to perform pixel-level classification of road areas, generating a confidence map of drivable road areas.
3. The human-like autonomous driving high-robustness control method based on visual reference points according to claim 1, characterized in that, Step S2 further includes: Based on the longitudinal coordinates of the target vehicle's bottom edge and the longitudinal reference point, construct the longitudinal proximity parameters and calculate the longitudinal rate of change; The lateral deviation parameters are constructed based on the lane center coordinates and the lateral reference point, and the lateral change rate is calculated. Common-mode suppression processing is performed on the longitudinal drift of the image based on the change in vehicle pitch angle, and a compensated longitudinal proximity parameter is constructed. The lane line coordinates are rotated and compensated based on the vehicle roll angle, and lateral compensation deviation parameters are constructed.
4. The human-like autonomous driving high-robustness control method based on visual reference points according to claim 1, characterized in that, Step S3 further includes: Based on the vehicle's current speed Constructing a dynamic longitudinal threshold: ; ; ; in, , , These represent the basic threshold parameters, , Indicates the speed adjustment coefficient; Based on longitudinal proximity parameters Constructing vertical semantic states: when When this occurs, it is determined to be a long-distance state; when At that time, it is determined to be a mid-range state; when At that time, it is determined to be a close-range situation; when When this occurs, it is determined to be a dangerous approach situation; The longitudinal proximity parameter is smoothed using a time-filtering model: ; in, Indicates the filter weight coefficients; The braking acceleration is constructed based on the smoothed longitudinal proximity parameters and the rate of change: ; in, Indicates the maximum braking acceleration. , This represents the control gain coefficient.
5. The human-like autonomous driving high-robustness control method based on visual reference points according to claim 1, characterized in that, Step S3 further includes: Constructing road curve curvature parameters using the rate of change of lane line curvature: ; in, Indicates the number of sampling points; Based on the curvature parameters Curve correction for longitudinal threshold: ; According to the vehicle pitch angle Correcting the longitudinal reference point: ; in, Indicates the pitch compensation coefficient; According to the curve rating Cooperative constraints on vehicle target speed: ; ; in, Indicates the base speed. Indicates the curve limit coefficient. Indicates the set cruise speed; According to the braking acceleration Target speed and the vehicle's current speed Construct target longitudinal control acceleration : ; in, Indicates the speed tracking gain coefficient; Based on the lateral compensation deviation parameter Horizontal deviation change rate and cornering level Construct the target steering wheel angle: ; in, This represents the lateral deviation control coefficient. This represents the horizontal dynamic compensation coefficient. This represents the cornering compensation coefficient.
6. The human-like autonomous driving high-robustness control method based on visual reference points according to claim 1, characterized in that, Step S4 further includes: The target acceleration generated in step S3 Steering wheel angle relative to the target The signals are output to the vehicle actuators and converted into throttle control signals, brake control signals, and steering control signals, respectively. Record vehicle historical trajectory points And by using extrinsic and intrinsic parameters from the camera, the historical trajectory points are projected onto the current image coordinate system: ; in, This represents the camera intrinsic parameter matrix. This represents the coordinate transformation matrix from the vehicle body to the camera; Construct a historical passable mask based on the projected historical trajectory area: ; And construct trajectory confidence weights using a time decay function: ; in, Indicates the initial weight parameters. Represents the time decay constant; Obtain the confidence map of the drivable region generated by semantic segmentation. The drivable area confidence map is then fused with the historical drivable mask to generate a fused drivable area confidence map. ; According to the curve rating Coordinated restrictions on the maximum permissible speed of vehicles: ; And according to the vehicle's set cruise speed Generate target vehicle speed: ; in, Indicates the basic vehicle speed parameter. Indicates the curve constraint coefficient; Based on the fused drivable area confidence map The confidence level of the vehicle’s current driving area is verified. When the confidence level of the corresponding area in the fused drivable area confidence map is lower than a preset threshold, deceleration control or braking control is triggered to improve the vehicle’s driving stability and robustness in scenarios with sudden changes in lighting, occlusion, and blurred road boundaries.
7. A robust human-like autonomous driving control system based on visual reference points, characterized in that: include: Image acquisition module, used to acquire image data of the road in front of the vehicle; The fixed reference point calibration module is used to acquire and store longitudinal and lateral reference points; The target detection module is used to detect target vehicles and output the target's bottom edge coordinate data; The lane line detection module is used to detect lane lines and output lane center coordinate data; The longitudinal and lateral deviation calculation module is used to calculate the longitudinal proximity parameter, the lateral deviation parameter, and their rate of change. The scene compensation module is used to compensate and correct longitudinal and lateral deviations based on the vehicle's pitch and roll angles. The longitudinal and lateral decision control module is used to generate target acceleration and target steering wheel angle based on dynamic thresholds, fuzzy rules, and finite state machines; The historical trajectory fusion module is used to construct a historical trajectory mask and fuse it with the semantic segmentation results; The collaborative control module is used to jointly constrain and control the vehicle's longitudinal speed and lateral steering according to the curve level. The execution module is used to output throttle, brake, and steering control signals.
8. The humanoid autonomous driving high-robustness control system based on visual reference points according to claim 7, characterized in that, The longitudinal and lateral deviation calculation module includes: The longitudinal proximity calculation unit is used to calculate the longitudinal proximity of the target vehicle. Lateral deviation calculation unit, used to calculate the lateral deviation of the lane center; The rate of change analysis unit is used to calculate the rate of change of longitudinal and transverse deviations and generate dynamic control trend parameters.
9. The humanoid autonomous driving high-robustness control system based on visual reference points according to claim 7, characterized in that, The historical trajectory fusion module includes: The trajectory recording unit is used to record the vehicle's historical trajectory points; The trajectory projection unit is used to map historical trajectories to the current image coordinate system using camera intrinsic and extrinsic parameters. The region fusion unit is used to generate a fused drivable region confidence map to improve the path robustness of vehicles in complex scenarios.
10. The humanoid autonomous driving high-robustness control system based on visual reference points according to claim 7, characterized in that, The collaborative control module includes: The curve recognition unit is used to identify the curve level based on road curvature parameters; Speed coordination control unit, used to generate the maximum permissible vehicle speed; The steering coordination control unit is used to adjust the steering parameters based on lateral deviation parameters, lateral change rate, and cornering grade. Output the target steering wheel angle.
