Vehicle air suspension leveling system based on a small number of height sensors and visual closed loop

By using a small number of height sensors and a vision closed-loop system, combined with visual pitch and roll values, three-degree-of-freedom leveling of the vehicle's air suspension is achieved. This solves the problems of a large number of sensors and poor stability in traditional systems, improves leveling accuracy and fault detection capabilities, and adapts to complex road environments.

CN123104306APending Publication Date: 2026-11-17ZHEJIANG VIE SCI & TECH +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611621432.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In traditional vehicle air suspension leveling systems, multiple height sensors increase material costs and assembly workload, and are susceptible to contamination, corrosion, and damage, resulting in poor stability. Furthermore, a single height measurement cannot distinguish between the average vehicle height, pitch, and roll, making accurate leveling control impossible.

Method used

Employing a small number of height sensors and a visual closed-loop system, the system combines a forward-facing camera, a vehicle height sensor, and an image processing module with a leveling controller. It utilizes visual pitch and roll values ​​to supplement the rotational degree of freedom information missing from the physical height sensor, achieving a three-degree-of-freedom leveling state. Furthermore, it improves leveling accuracy and diagnostic capabilities through cross-identification of visual differential response and physical vertical response, as well as separation of road slope.

Benefits of technology

While reducing the number of sensors, it maintains or improves leveling accuracy and diagnostic capabilities, reduces costs and failure risks, adapts to different road scenarios, avoids misjudging road slope as suspension error, and achieves online load compensation and fault detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123104306A_ABST
    Figure CN123104306A_ABST
Patent Text Reader

Abstract

This invention belongs to the field of vehicle air suspension leveling technology, and relates to a vehicle air suspension leveling system based on a small number of height sensors and a visual closed loop, solving the problems of high cost due to numerous height sensors and difficulty in observing three degrees of freedom after reducing the number of sensors. The system includes a forward-facing camera, a small number of vehicle height sensors, an image processing module, and a leveling controller. The forward-facing camera is an existing ADAS camera. The image processing module determines the vehicle pitch and roll values ​​from the forward image. The leveling controller uses the physical height value and the visual pitch and roll values ​​to form the leveling state, and controls the air suspension components in a closed loop. When stationary or in low dynamic conditions, the leveling controller applies inflation / deflation commands, and uses visual differential response and physical vertical response to cross-identify, detect the control valve, and verify the visual signal. The image processing module outputs the lane line surface normal and the horizon gravity reference direction in parallel, separating the road slope and vehicle attitude deviation, realizing three-degree-of-freedom leveling, online diagnosis, and slope adaptation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of vehicle air suspension leveling technology, and specifically relates to a vehicle air suspension leveling system based on a small number of height sensors and a visual closed loop. Background Technology

[0002] Trucks, buses, coaches, motorhomes and other heavy vehicles typically use air springs to maintain vehicle height and posture in response to changes in occupant distribution, cargo load, fuel level, kneeling command, coupling operation or axle load conditions.

[0003] Traditional electronic leveling systems typically place mechanical or electromechanical vehicle height sensors at multiple suspension locations, such as one sensor at each independently controlled corner point. The sensor signals are converted into inflation and deflation commands for the corresponding air springs.

[0004] However, the addition of vehicle height sensors, brackets, linkages, wiring harnesses, connectors, calibration work, and diagnostic coverage increases material costs and assembly workload. Furthermore, exposed height sensors are susceptible to contamination, corrosion, damage, tolerance accumulation, and misadjustment during maintenance. The more height sensors there are, the worse the stability becomes. However, relying solely on a single height measurement cannot distinguish between the average vehicle height, pitch, and roll, thus failing to achieve the intended measurement and leveling control. Summary of the Invention

[0005] The purpose of this invention is to address the aforementioned problems in the existing technology by proposing a vehicle air suspension leveling system based on a small number of height sensors and a visual closed loop.

[0006] The objectives of this invention can be achieved through the following technical solutions:

[0007] A vehicle air suspension leveling system based on a small number of height sensors and a visual closed loop includes a vehicle, the vehicle comprising a vehicle body and a sprung body supported on the vehicle body by multiple controllable air suspension elements, the vehicle air suspension leveling system comprising:

[0008] A forward-facing camera, which is fixed to the sprung vehicle body and calibrated relative to the vehicle coordinate system, is capable of acquiring forward image data.

[0009] The vehicle height sensor, located between the vehicle body and the sprung body, is fewer in number than the air suspension components and can detect the distance between the vehicle body and the sprung body.

[0010] The image processing module is able to determine the pitch and roll values ​​of the sprung vehicle body relative to a reference attitude based on the image data collected by the forward-facing camera.

[0011] The leveling controller can control the air suspension components based on the height value obtained by the vehicle height sensor and the leveling state represented by the pitch and roll values ​​determined by the image processing module, so that the vehicle body attitude reaches the target state.

[0012] When the vehicle is stationary or in a low dynamic state, the leveling controller applies short-term inflation or deflation commands to the selected air suspension components. Based on the pitch angle and roll angle differential response measured by the forward-facing camera and the vertical response measured by the vehicle height sensor, it identifies or updates the actuator geometry matrix, detects the control valves of air suspension components with insufficient performance, and verifies whether the sign and amplitude of the visual attitude signal meet expectations.

[0013] The image processing module outputs the road surface plane normal based on the lane lines and the gravity reference direction based on the horizon in parallel. When the deviation between the two exceeds the reasonable slope range, the leveling controller determines it as road slope and maintains the calibration attitude relative to the road in the road relative mode. When the deviation is within the suspension error range, it determines it as vehicle body attitude deviation and performs closed-loop correction.

[0014] The forward-facing camera is an existing ADAS forward-facing camera in the vehicle that has been calibrated with internal and external parameters. Its image data is transmitted to the image processing module via the vehicle network bus. The system does not have a dedicated suspension attitude camera.

[0015] In the vehicle air suspension leveling system of this invention, a forward-facing camera is rigidly connected to the sprung vehicle body and completes the vehicle coordinate system calibration, enabling it to perceive the attitude changes of the vehicle body relative to the road scene. The number of vehicle height sensors is intentionally set to be less than the number of air suspension components. Physical height sensors are no longer arranged at each independently controlled suspension position, but only a small number of physical height sensors provide absolute height or vertical translation reference. The image processing module extracts the pitch and roll values ​​of the sprung vehicle body relative to the reference attitude from the forward image, visually supplementing the two rotational degrees of freedom information missing by the reduced physical height sensors. The leveling controller combines the vertical translation information represented by the physical height value with the rotational information represented by the visual pitch and roll values ​​to form a complete leveling state, and controls each air suspension component in a closed loop accordingly. The states include pitch and roll angles less than thresholds, or adapted to road slope / lateral tilt, avoiding misjudging road longitudinal or transverse slopes as suspension errors and improving the applicability of the system on slopes and inclined roads. In other words, the physical height value provides an absolute vertical reference, while visual estimation provides the two rotational degrees of freedom needed to determine the vehicle's plane. Thus, even with fewer physical height sensors than air suspension components, it still achieves the leveling state of three degrees of freedom—vehicle height, pitch, and roll—and realizes closed-loop leveling. Compared to existing technologies that place vehicle height sensors at each suspension location or use cameras only for predictive suspension control or road environment recognition, this solution achieves a synergistic effect through the complementary relationship between the sensor quantity architecture and the degrees of freedom of visual attitude feedback. This reduces the number of height sensors, brackets, links, wiring harnesses, and connectors, lowering costs, assembly workload, and failure risks. Furthermore, the visual closed-loop allows the leveling controller to continuously observe the vehicle's actual response to the air spring's movement, maintaining or even improving leveling accuracy and diagnostic capabilities while reducing the number of physical sensors. Meanwhile, by using short-term inflation or deflation excitation under static or low-dynamic conditions, the visual attitude differential response and physical vertical response are cross-identified and verified, enabling the visual channel and the actuator physical channel to mutually verify each other, thereby improving long-term operational accuracy, fault diagnosis capability, and calibration drift compensation capability. By using parallel output and comparison based on the lane line plane normal and the horizon gravity reference direction, the longitudinal slope and transverse slope of the road and the vehicle body's own attitude deviation can be estimated separately, avoiding mistaking road slope for suspension error.

[0016] In short, the distinguishing technical features of this solution include at least the following: the number of vehicle height sensors is less than that of air suspension components, providing an absolute vertical translation reference based on physical height values; the leveling controller combines physical height values ​​with visual pitch and roll values ​​to form a three-degree-of-freedom leveling state; controlled inflation / deflation excitation is applied when stationary or in low-dynamic conditions, using visual differential response and physical vertical response to cross-identify, detect, and verify the visual signal symbol and amplitude; and parallel output of the road surface plane normal based on lane lines and the gravity reference direction based on the horizon, separating and estimating road slope and vehicle attitude deviation. This solution addresses the problems of missing absolute vertical reference after sensor reduction, unobservable three-degree-of-freedom states, lack of cross-validation between the visual channel and the actuator physical channel, and the easy confusion between road slope and vehicle attitude deviation.

[0017] In particular, the cross-identification and verification of controlled charging / exhausting excitation and visual differential response and physical vertical response under vehicle stationary or low dynamic conditions enables the visual attitude channel and the physical channel of the air suspension actuator to mutually verify each other. This allows for online updates of the actuator geometry matrix, compensation of load-related aerodynamic characteristics, detection of control valves with insufficient performance, and verification of the sign and amplitude of the visual attitude signal. Based on the parallel comparison and slope separation of the lane line plane normal and the gravity reference direction based on the horizon, the longitudinal slope and transverse slope of the road and the attitude deviation of the vehicle body relative to the road can be estimated separately, and the system can switch between road relative mode and gravity relative stationary mode to avoid mistaking road slope for suspension error.

[0018] In the aforementioned vehicle air suspension leveling system based on a small number of height sensors and a visual closed loop, the image processing module includes a visual attitude estimator configured to select from a variety of complementary attitude estimation methods to determine pitch and roll values. The attitude estimation methods include:

[0019] The lane-line-based method obtains at least one of longitudinal vanishing point, apparent convergence, perspective scale, and left-right asymmetry from the lane lines in the forward image data, and determines the pitch and / or roll values ​​accordingly.

[0020] The horizon-based method detects the horizon in the foreground image data, determines the yaw value based on the tilt angle of the horizon, and determines the pitch value based on the calibrated vertical image position.

[0021] And an alternative method for situations where there are no lane lines and / or no horizon, which obtains the roll value through edge detection and robust statistics on the edges of the near-horizontal and vertical scene, and obtains one or more vanishing points through the intersections or convergences of selected families of lines, so as to estimate the pitch value based on the one or more vanishing points;

[0022] The leveling controller determines whether the visual feedback is effective based on the confidence level output by the visual attitude estimator, and controls the air suspension components based on the pitch and roll values ​​determined by the image processing module when the visual feedback is effective.

[0023] The visual pose estimator integrates multiple pose estimation paths and can select or switch them according to scene conditions. It also outputs confidence scores for the leveling controller to determine whether the visual feedback is usable. As a result, the system can obtain usable pitch and roll values ​​in different road scenarios, such as when lane lines are clear, the horizon is visible, or there are no lane lines or horizons. This improves the environmental adaptability and robustness of visual closed-loop leveling and avoids misusing unreliable pose estimations when visual conditions are poor.

[0024] In the aforementioned vehicle air suspension leveling system based on a small number of height sensors and a visual closed loop, the lane line-based method includes: when a longitudinal lane line or road boundary is detected, fitting left and right lane models and determining their convergence in the image;

[0025] The roll value is determined based on the lane geometric rotation caused by roll changes and the asymmetric changes in the apparent lane width, slope, or perspective scale on both sides of the image. The pitch value is determined based on the vertical displacement of the longitudinal vanishing point caused by pitch changes and the apparent convergence and scale change patterns of lane lines.

[0026] The measured features are compared with the features predicted based on camera calibration and vehicle nominal attitude.

[0027] Furthermore, lane measurements are accumulated across multiple image frames and robustly fitted using RANSAC, M-estimation, or Kalman filtering.

[0028] For curved lanes, the lane model is represented by a polynomial or spiral and local extrapolation is performed to obtain the vanishing direction; based on the road model and uncertainty estimation, it is not required that the boundaries of the two lanes be completely parallel in the image.

[0029] This invention utilizes lane line geometry and camera calibration parameters to establish an attitude observation model. Robust estimation is achieved through left and right lane model fitting, vanishing point determination, and multi-frame accumulation. Curved lanes can be described using polynomials or spiral curves. Therefore, in road environments with lane lines, it can estimate vehicle pitch and roll with high accuracy, exhibiting good adaptability to imperfectly parallel lane boundaries, curved lanes, and measurement anomalies, thus improving the reliability of visual leveling in real-world road conditions.

[0030] In the above-mentioned vehicle air suspension leveling system based on a small number of height sensors and visual closed loop, the horizon-based method includes: identifying the horizon formed by the skyline, road and sky boundary, distant terrain boundary, and building line through a horizon detector, or inferring the horizon statistically based on horizontal texture and motion characteristics.

[0031] The tilt value is determined by the inclination of the horizon relative to the x-axis of the image, and the pitch value is determined by its vertical position in combination with the camera's intrinsic parameters and nominal calibration.

[0032] When the horizon is obscured, is significantly nonlinear, or inconsistent with other visual features, the corresponding estimate is rejected.

[0033] This invention maps the horizon tilt and vertical position to lateral and pitch observations respectively using a horizon detector, and rejects unreliable horizons. As a result, the vehicle attitude can still be obtained in scenarios where there are no lane lines but a visible horizon exists. By rejecting horizon estimates that are occluded, nonlinear, or inconsistent with other features, erroneous leveling is avoided, and the robustness of the system in open roads and complex environments is improved.

[0034] In the above-mentioned vehicle air suspension leveling system based on a small number of height sensors and visual closed loop, the alternative methods for situations without lane lines and / or without a horizon include: performing edge detection, line segment detection, or learning-based structural feature extraction.

[0035] Line segments are selected based on their length, contrast, duration, semantic category, and proximity to the horizontal or vertical direction after compensation based on the current pose assumption; an edge direction histogram or probability distribution is formed, and the dominant rotation of the near-horizontal and near-vertical line families is used as the tilt value.

[0036] Based on the vertical image position of the vanishing point, the convergence or divergence of near-vertical scene edges, the perspective scale of repeating structures, or the pitch value estimated from the road surface plane obtained by optical flow or monocular depth, multiple vanishing points are used in the structured scene geometry model to exclude dynamic objects through semantic segmentation, motion inconsistency, or multi-frame tracking.

[0037] This invention establishes backup pose observations using features such as edges, line segments, vanishing points, optical flow, or monocular depth in the scene when lane lines and horizons are absent. Dynamic objects are excluded through semantic segmentation and multi-frame tracking, thereby expanding the applicable scenarios for visual pose estimation. It can continue to obtain usable tilt and pitch information in urban environments, structured roads, or complex backgrounds, improving the usability and estimation stability of visual closed-loop leveling in lane-line-less environments. The structured scene geometric model includes the Manhattan world model or the Atlanta world model. World models based on orthogonal or quasi-orthogonal direction assumptions describe multiple vanishing points in the scene, improving the stability of vanishing point detection and pitch angle estimation, and giving the backup method stronger geometric constraints and anti-interference capabilities in structured scenes.

[0038] In the above-mentioned vehicle air suspension leveling system based on a small number of height sensors and visual closed loop, the leveling controller determines whether the visual feedback is effective based on the confidence level, including that each visual method provides an estimate and a confidence level, and the confidence level depends on at least one of the following: the number and distribution of features, fitting residuals, inter-frame consistency, image exposure, visibility, weather, camera occlusion, road curvature, and consistency between independent methods.

[0039] The image processing module also includes a fusion module, which effectively observes by combining weighted averaging, filtering, Bayesian estimation, or learning models. The fusion module continuously modulates the fusion gain of visual observations in the leveling state estimation with confidence level, and the lower the confidence level, the higher the weight of the physical height channel. During the transition period when the confidence level crosses the threshold, a rate limit is applied to the fusion gain to suppress leveling state jumps. When the confidence level is lower than the threshold, visual closed-loop leveling is paused, and the system maintains the last safe state of each valve of the air suspension components, or enters a degraded function mode based on the remaining vehicle height sensor.

[0040] This invention incorporates a confidence assessment and fusion module, which weights and combines estimates and confidence levels from multiple visual methods. The confidence level is then linked to a safety degradation strategy to prevent erroneous leveling commands from being issued under poor visual conditions or when estimates are unreliable, thus improving system safety and robustness. Furthermore, even when the visual channel is unavailable, the remaining vehicle height sensors can maintain basic leveling or allow the system to enter a safe state. Further, by continuously modulating the fusion gain with the confidence level and applying a rate limit during the threshold transition period, the abrupt changes in the leveling state during visual channel failure or recovery can be suppressed, resulting in a smoother weight switching between the visual channel and the physical height channel.

[0041] In the aforementioned vehicle air suspension leveling system based on a small number of height sensors and a visual closed loop, the vehicle height sensor is located at the known vehicle plane coordinates (x... s , y s The height z measured at point ) s As a vertical translation reference, based on the pitch value θ and tilt value φ determined by the image processing module, for coordinates (x... i , y i The suspension position is used to calculate the virtual vehicle height z. i :

[0042] z i = z s + (x i - x s ) tan(θ) + (y i - y s ) tan(φ),

[0043] The calculated virtual vehicle height zi As a virtual corner height signal for the leveling controller;

[0044] The vehicle height sensor consists of three sensors, which are respectively located at the front left, front right, and rear axle positions of the vehicle. The leveling controller performs consistency verification between the pitch / roll angle reconstructed from the height difference between any two vehicle height sensors and the pitch / roll value output by the image processing module. When the deviation continues to exceed the limit, the camera extrinsic parameter calibration is updated to compensate for calibration drift.

[0045] The physical height value provided by the vehicle height sensor is used to provide vertical translation information, and the pitch and roll values ​​provided by the image processing module are used to provide the two rotational degrees of freedom required to determine the vehicle plane.

[0046] This invention reconstructs the virtual height of multiple suspension positions using a physical height reference and two visual angle measurements. Geometric calculations are performed using known sensor coordinates and suspension position coordinates, achieving multi-position height estimation with fewer physical sensors. This allows the leveling controller to obtain signals from multiple vehicle height sensors to control each air suspension component, maintaining three-degree-of-freedom leveling capability while reducing the number of sensors and lowering system cost and complexity. Through differential reconstruction from three vehicle height sensors and consistency verification of visual pitch / roll values, redundancy verification and camera extrinsic parameter calibration drift compensation are achieved within the sensor reduction architecture, improving long-term system reliability.

[0047] In the above-mentioned vehicle air suspension leveling system based on a small number of height sensors and visual closed loop, the leveling controller is configured with a road relative mode and a gravity relative stationary mode.

[0048] In the road-relative mode, the image processing module determines the orientation of the forward-facing camera relative to the locally observed road surface plane, and the leveling controller keeps the sprung vehicle body in its calibrated attitude relative to the road.

[0049] In the gravity-relative static mode, the leveling controller compares the road surface plane orientation based on lane lines with the orientation based on the horizon to separate the road longitudinal and transverse slopes from the attitude of the sprung vehicle body relative to the road.

[0050] The leveling controller is allowed to issue a leveling command only when quasi-static conditions are met. The quasi-static conditions include: vehicle speed is below a threshold; parking brake or service brake has been applied; or the vehicle is determined to be in a quasi-static state based on at least one of longitudinal acceleration, lateral acceleration, yaw rate, steering angle, wheel speed change and image motion.

[0051] During driving, the leveling controller employs low-pass filtering, time window, and dynamic compensation, and eliminates transient pitch during braking and transient roll during turning from static height correction;

[0052] When the vehicle is in motion, the pitch and roll values ​​determined by the image processing module are used as closed-loop feedback quantities, and the height value obtained by the vehicle height sensor is used as an auxiliary feedback quantity; when the vehicle is stationary or at low speed, the height value is used as the main feedback quantity, and attitude correction is performed based on the pitch and roll values.

[0053] This invention sets up a road relative mode, a gravity relative static mode, quasi-static judgment conditions, and different feedback quantity allocation strategies for driving and stationary conditions. By comparing the direction separation of road slope and vehicle posture based on lane lines and horizon, it can distinguish between road longitudinal slope, cross slope and the vehicle's own posture deviation, avoid mistaking road slope for suspension error, and prevent incorrect leveling under dynamic conditions such as braking and turning, thereby improving the accuracy, safety and adaptability of leveling control.

[0054] In the above-mentioned vehicle air suspension leveling system based on a small number of height sensors and visual closed loop, the ADAS forward-facing camera is one or more of a single forward-facing camera, a stereo camera, a fisheye camera, or a multi-camera group. The forward-facing camera is fixed on the inside of the windshield or the front of the roof of the sprung vehicle body and has internal and external parameters calibrated relative to the vehicle coordinate system.

[0055] The vehicle height sensor includes a linkage sensor, an axle position sensor, or a pressure / load fusion model verified by a physical sensor;

[0056] The pitch and roll values ​​determined by the image processing module can be used for the leveling control of the leveling controller, and can also be used to detect camera calibration drift caused by changes in suspension or load.

[0057] The leveling controller includes a deviation determination unit, a control quantity generation unit, and a drive unit. The deviation determination unit determines the deviation between the current vehicle body posture and the target state based on the height value, pitch value, and roll value. The control quantity generation unit generates the target height, target pressure, or target gas quantity for each air suspension element based on the deviation. The drive unit controls the inflation, deflation, or pressure maintenance of the air suspension elements based on the target height, target pressure, or target gas quantity.

[0058] The visual attitude information of this invention serves both leveling control and ADAS camera calibration drift detection. This means it can reuse existing ADAS cameras and vehicle networks to achieve closed-loop air suspension leveling and promptly detect camera calibration drift caused by suspension or load changes, improving the collaborative reliability of the vehicle's perception system and suspension system. The forward-facing camera is an existing ADAS forward-facing camera calibrated with intrinsic and extrinsic parameters. Its image data is transmitted to the image processing module via the vehicle network bus. The system does not require a dedicated suspension attitude camera, thus achieving visual closed-loop leveling without adding a dedicated suspension attitude camera.

[0059] As an optimization, the leveling controller is configured with different driving heights, different loading heights, bus kneeling descent, trailer connection, docking, ferry, maintenance, or static leveling states. This allows the system to adapt to the vehicle's posture requirements in different usage scenarios, improving the functionality and practicality of the air suspension leveling system.

[0060] As an optimization, the vehicle height sensor includes a main sensor and at least one redundant sensor, and the vehicle height sensor is located at the center or one side of the axle; this can improve the reliability of height measurement while maintaining a reduction in the number of sensors, and support degraded operation or rationality verification when the main sensor is abnormal.

[0061] As an optimization, the vehicle height sensor includes three vehicle height sensors, which are respectively arranged at the left front, right front and rear axle positions of the vehicle, and the number of vehicle height sensors is less than the number of controllable air suspension components; it can obtain the vehicle's height information in the front-rear and left-right directions with fewer physical sensors, and cooperate with visual pitch and roll values ​​to achieve reliable vehicle attitude estimation and leveling control.

[0062] As an optimization, the air suspension components include air springs, solenoid valves connected to the air springs, compressors, air tanks, and exhaust valves. The leveling controller can precisely adjust the amount of gas in each air spring through inflation, deflation, and pressure holding, thereby achieving closed-loop control of vehicle height, pitch, and roll.

[0063] Compared with the prior art, the present invention has the following main advantages:

[0064] 1. By using fewer physical height sensors than air suspension components in conjunction with forward-looking visual attitude estimation, three-degree-of-freedom leveling of vehicle height, pitch, and roll is achieved, reducing the cost of sensors, brackets, wiring harnesses, and assembly.

[0065] 2. Reusing the vehicle's existing calibrated forward-facing camera and vehicle network eliminates the need for additional dedicated suspension cameras, reducing hardware complexity and the difficulty of vehicle integration.

[0066] 3. By using physical height values ​​to provide an absolute vertical reference, visual pitch and roll values ​​to provide two degrees of rotational freedom, and by calculating and reconstructing the height of each suspension position through virtual corner height, the observability problem after sensor reduction is solved.

[0067] 4. By using multiple visual methods such as lane lines, horizon, and edge / vanishing point, and confidence-weighted fusion, it adapts to different road scenarios and pauses leveling or degrades operation when visual perception is unreliable, thereby improving safety and robustness.

[0068] 5. By leveraging visual closed-loop and controlled excitation, online identification, load aerodynamic compensation, fault diagnosis, and ADAS camera calibration drift detection are achieved, and the errors of road longitudinal slope, cross slope, and vehicle posture are distinguished to avoid incorrect leveling.

[0069] 6. By cross-identifying the visual pitch / tilt differential response with the physical vertical response under controlled charging / discharging excitation in static or low-dynamic states, the system can detect underperforming control valves and verify the visual attitude signal symbol and amplitude, thereby improving the system's long-term diagnostic capabilities.

[0070] 7. By comparing the road surface plane normal and the gravity reference direction based on the horizon in parallel, the road slope and vehicle attitude deviation can be estimated separately, and the leveling accuracy under slope and inclined road surface can be improved by switching between road relative mode and gravity relative static mode.

[0071] The aforementioned advantages further demonstrate that, based on "a physical height sensor with fewer components than air suspension components + visual attitude closed loop", this invention further introduces two linked features: controlled excitation cross-identification verification and dual visual cue slope separation and leveling. This allows the visual attitude channel and the physical channel of the air suspension actuator to mutually verify each other, enabling the road slope and vehicle posture deviation to be estimated separately. Thus, while reducing the number of vehicle height sensors, it still maintains three-degree-of-freedom leveling capability, online diagnostic capability, calibration drift compensation capability, and slope adaptability. Attached Figure Description

[0072] Figure 1 This is a schematic diagram of the arrangement of the vehicle and its components provided by the present invention;

[0073] Figure 2 This is a schematic diagram of the closed-loop adjustment architecture of the vehicle air suspension leveling system provided by the present invention;

[0074] Figure 3 This is an image pose feature map provided by the present invention;

[0075] Figure 4 This is a schematic diagram of the leveling process provided by the present invention.

[0076] In the diagram, 1 is the vehicle, 2 is the vehicle body, 3 is the air suspension components, 4 is the sprung body, 5 is the forward-facing camera, and 6 is the vehicle height sensor. Detailed Implementation

[0077] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0078] Specific implementation examples Figures 1-4 As shown, this vehicle air suspension leveling system based on a small number of height sensors and visual closed loop includes a vehicle 1, which includes a vehicle body 2 and a sprung body 4 supported on the vehicle body 2 by multiple controllable air suspension elements 3. The vehicle air suspension leveling system includes:

[0079] The forward-facing camera 5 is fixed to the sprung body 4 and calibrated relative to the vehicle coordinate system, and can collect forward image data.

[0080] The vehicle height sensor 6 is located between the vehicle body 2 and the sprung body 4, and its number is less than that of the air suspension components 3. It can detect the distance between the vehicle body 2 and the sprung body 4.

[0081] The image processing module is able to determine the pitch and roll values ​​of the sprung body 4 relative to a reference attitude based on the image data collected by the forward-facing camera 5.

[0082] The leveling controller can control the air suspension element 3 to achieve the target vehicle posture based on the height value obtained by the vehicle height sensor 6 and the leveling state represented by the pitch and roll values ​​determined by the image processing module.

[0083] When vehicle 1 is stationary or in a low-dynamic state, the leveling controller applies a short-term inflation or deflation command to the selected air suspension element 3. The forward-facing camera 5 measures the pitch and roll angle differential responses generated by the command, and the leveling controller obtains the corresponding vertical response through the vehicle height sensor 6. Based on the pitch angle differential response, roll angle differential response, and vertical response, the leveling controller identifies or updates the actuator geometry matrix, compensates for load-related aerodynamic characteristics, detects the control valves of underperforming air suspension elements 3, and verifies whether the sign and amplitude of the visual attitude signal meet expectations.

[0084] The image processing module outputs the road surface plane normal based on the lane lines and the gravity reference direction based on the horizon in parallel. When the deviation between the two exceeds the reasonable slope range, the leveling controller determines it as road slope and maintains the calibration attitude relative to the road in the road relative mode. When the deviation is within the suspension error range, it determines it as vehicle body attitude deviation and performs closed-loop correction.

[0085] The forward-facing camera 5 is an existing ADAS forward-facing camera in vehicle 1 that has been calibrated with internal and external parameters. Its image data is transmitted to the image processing module via the vehicle network bus. The system does not have a dedicated suspension attitude camera.

[0086] Specifically, in the vehicle air suspension leveling system of the present invention, the forward-facing camera 5 is rigidly connected to the sprung body 4 and completes the vehicle coordinate system calibration, enabling it to perceive the attitude changes of the body relative to the road scene; the number of body height sensors 6 is intentionally set to be less than the number of air suspension elements 3, and physical height sensors 6 are no longer arranged at each independently controlled suspension position, but only a small number of physical height sensors 6 provide absolute height or vertical translation reference; the image processing module extracts the pitch and roll values ​​of the sprung body 4 relative to the reference attitude from the forward image, visually supplementing the two missing rotational degrees of freedom information of the reduced physical height sensors 6; the leveling controller combines the vertical translation information represented by the physical height value with the rotational information represented by the visual pitch and roll values ​​to form a complete leveling state, and controls each air suspension element 3 in a closed loop accordingly, wherein the state includes pitch and roll angles less than the threshold, or adapted to the road slope / lateral tilt, avoiding misjudging the longitudinal or transverse slope of the road as suspension error, and improving the applicability of the system on slopes and inclined roads. In other words, the physical height value provides an absolute vertical reference, while visual estimation provides the two rotational degrees of freedom needed to determine the vehicle's plane. Thus, even with fewer physical height sensors (6) than air suspension components (3), it still achieves leveling in three degrees of freedom—body height, pitch, and roll—and realizes closed-loop leveling. This reduces the number of height sensors, brackets, links, wiring harnesses, and connectors, lowering costs, assembly workload, and failure risks. Furthermore, the visual closed-loop allows the leveling controller to continuously observe the vehicle's actual response to the air spring's movement, maintaining or even improving leveling accuracy and diagnostic capabilities while reducing the number of physical sensors (6). Moreover, through cross-identification of visual differential response and physical vertical response under controlled excitation, and slope separation based on lane lines and horizon dual visual cues, the system's online diagnostics, drift compensation calibration, and slope leveling adaptability can be further improved.

[0087] In this embodiment, the image processing module includes a visual attitude estimator configured to select from a variety of complementary attitude estimation methods to determine pitch and roll values. The attitude estimation methods include lane-line-based methods, horizon-based methods, and alternative methods for situations without lane lines and / or without a horizon.

[0088] First, the lane-line-based method obtains at least one of the following from the lane lines in the preceding image data: longitudinal vanishing point, apparent convergence, perspective scale, and left-right asymmetry, and determines the pitch and / or roll values ​​accordingly. Specifically, when a longitudinal lane line or road boundary is detected, the method fits left and right lane models and determines their convergence in the image; the roll value is determined based on the lane geometric rotation caused by roll changes and the asymmetric changes in apparent lane width, slope, or perspective scale on both sides of the image; the pitch value is determined based on the vertical displacement of the longitudinal vanishing point caused by pitch changes and the apparent convergence and scale change patterns of the lane lines; the measured features are compared with features predicted based on camera calibration and vehicle nominal attitude; lane measurements are accumulated across multiple image frames and robustly fitted using RANSAC, M-estimation, or Kalman filtering; for curved lanes, the lane model is represented using a polynomial or spiral curve and local extrapolation is performed to obtain the vanishing direction; based on the road model and uncertainty estimation, it is not required that the two lane boundaries be completely parallel in the image.

[0089] This method utilizes lane line geometry and camera calibration parameters to establish an attitude observation model. Robust estimation is achieved through left and right lane model fitting, vanishing point determination, and multi-frame accumulation. Curved lanes can be described using polynomials or spiral curves. Therefore, it can estimate vehicle pitch and roll with high accuracy in road environments with lane lines, and exhibits good adaptability to imperfectly parallel lane boundaries, curved lanes, and measurement anomalies, improving the reliability of visual leveling in real-world road conditions.

[0090] Secondly, the horizon-based method detects the horizon in the foreground image data, determines the lateral tilt value based on the tilt angle of the horizon, and determines the pitch value based on the calibrated vertical image position. Specifically, this includes: identifying the horizon formed by the skyline, road and sky boundaries, distant terrain boundaries, and building lines through a horizon detector, or the horizon statistically inferred from horizontal texture and motion features; determining the lateral tilt value based on the tilt of the horizon relative to the image x-axis, and determining the pitch value based on its vertical position combined with camera intrinsic parameters and nominal calibration; rejecting the corresponding estimate when the horizon is occluded, significantly nonlinear, or inconsistent with other visual features.

[0091] This leveling system maps the horizon tilt and vertical position to roll and pitch observations respectively using a horizon detector, and rejects unreliable horizons. As a result, it can still obtain the vehicle attitude in scenarios without lane lines but with a visible horizon. By rejecting horizon estimates that are occluded, nonlinear, or inconsistent with other features, it avoids incorrect leveling and improves the robustness of the system in open roads and complex environments.

[0092] Another alternative method is used when there are no lane lines and / or no horizon. This method obtains tilt values ​​through edge detection and robust statistics on near-horizontal and vertical scene edges, and obtains one or more vanishing points by selecting the intersections or convergences of line families to estimate pitch values ​​based on these vanishing points. Specifically, this includes: performing edge detection, line segment detection, or learned structural feature extraction; filtering line segments based on line length, contrast, temporal persistence, semantic category, and proximity to the horizontal or vertical direction after compensation for the current pose assumption; forming an edge orientation histogram or probability distribution, using the dominant rotation of near-horizontal and near-vertical line families as tilt values; estimating pitch values ​​based on the vertical image position of the longitudinal vanishing point, the convergence or divergence of near-vertical scene edges, the perspective scale of repeating structures, or the road surface plane obtained from optical flow or monocular depth; and using multiple vanishing points in the structured scene geometry model to exclude dynamic objects through semantic segmentation, motion inconsistencies, or multi-frame tracking.

[0093] When lane lines and horizons are missing, this leveling system utilizes features such as edges, line segments, vanishing points, optical flow, or monocular depth in the scene to establish backup pose observations. It also excludes dynamic objects through semantic segmentation and multi-frame tracking, thereby expanding the applicable scenarios for visual pose estimation. This allows for the acquisition of usable tilt and pitch information in urban environments, structured roads, or complex backgrounds, improving the usability and estimation stability of visual closed-loop leveling in lane-line-less environments. The structured scene geometric models include the Manhattan world model or the Atlanta world model. These world models, based on orthogonal or quasi-orthogonal direction assumptions, describe multiple vanishing points in the scene, improving the stability of vanishing point detection and pitch angle estimation. This gives the backup method stronger geometric constraints and anti-interference capabilities in structured scenes.

[0094] Furthermore, the leveling controller determines the validity of visual feedback based on the confidence level output by the visual attitude estimator, and controls the air suspension element 3 based on the pitch and roll values ​​determined by the image processing module when the visual feedback is valid. In other words, the visual attitude estimator integrates multiple attitude estimation paths and can select or switch them according to scene conditions, while outputting a confidence level for the leveling controller to determine whether the visual feedback is usable. Thus, the system can obtain usable pitch and roll values ​​in different road scenarios, such as when lane lines are clear, the horizon is visible, or there are no lane lines or horizons, improving the environmental adaptability and robustness of visual closed-loop leveling and avoiding the misuse of unreliable attitude estimations when visual conditions are poor.

[0095] In this embodiment, the leveling controller determines the effectiveness of visual feedback based on confidence level. Each visual method provides an estimate and confidence level, which depends on at least one of the following: the number and distribution of features, fitting residuals, inter-frame consistency, image exposure, visibility, weather, camera occlusion, road curvature, and consistency between independent methods. The image processing module also includes a fusion module, which combines effective observations through weighted averaging, filtering, Bayesian estimation, or a learning model. The fusion module continuously modulates the fusion gain of visual observations in the leveling state estimation based on confidence level, with higher weights for the physical height channel as confidence level decreases. A rate limit is applied to the fusion gain during the transition period when the confidence level crosses a threshold to suppress leveling state transitions. When the confidence level is below the threshold, visual closed-loop leveling is paused, and the system maintains the final safe state of each valve of the air suspension element 3, or enters a degraded function mode based on the remaining vehicle height sensor 6.

[0096] Specifically, this invention includes a confidence assessment and fusion module that weights and combines estimates and confidence levels from multiple visual methods. The confidence level is then linked to a safety degradation strategy to prevent erroneous leveling commands from being issued under poor visual conditions or when estimates are unreliable, thus improving system safety and robustness. Furthermore, even when the visual channel is unavailable, the remaining vehicle height sensor 6 can maintain basic leveling or enter a safe state. By continuously modulating the fusion gain and limiting the transition rate, the weight switching between the visual channel and the physical height channel is made smoother, preventing abrupt changes in the leveling state caused by sudden shifts in confidence levels.

[0097] In this embodiment, the vehicle height sensor 6 is located at the known vehicle plane coordinates (x... s , y s The height z measured at point ) s As a vertical translation reference, based on the pitch value θ and tilt value φ determined by the image processing module, for coordinates (x... i , y i The suspension position is used to calculate the virtual vehicle height z. i :

[0098] z i = z s + (x i - x s ) tan(θ) + (y i - y s tan(φ)

[0099] The calculated virtual vehicle height z i As a virtual corner height signal for the leveling controller;

[0100] There are three vehicle height sensors 6, which are respectively located at the left front, right front and rear axle positions of vehicle 1; the leveling controller will verify the consistency between the pitch / roll angle reconstructed by the height difference between any two vehicle height sensors 6 and the pitch / roll value output by the image processing module. When the deviation continues to exceed the limit, the camera extrinsic parameter calibration will be updated to compensate for calibration drift.

[0101] The physical height value provided by the vehicle height sensor 6 is used to provide vertical translation information, while the pitch and roll values ​​provided by the image processing module are used to provide the two rotational degrees of freedom required to determine the vehicle's plane.

[0102] Specifically, this invention reconstructs the virtual height of multiple suspension positions using a physical height reference and two visual angle measurements. Geometric calculations are performed using known sensor coordinates and suspension position coordinates, achieving multi-position height estimation with fewer physical sensors. This allows the leveling controller to obtain signals from multiple vehicle height sensors to control each air suspension element 3, maintaining three-degree-of-freedom leveling capability while reducing the number of sensors and lowering system cost and complexity. Through differential reconstruction from three vehicle height sensors and consistency verification of visual angles, redundancy verification and online compensation for camera extrinsic parameter calibration drift are achieved within the sensor reduction architecture.

[0103] As an optimization of this embodiment, the leveling controller is configured with a road-relative mode and a gravity-relative static mode. In the road-relative mode, the image processing module determines the orientation of the forward-facing camera 5 relative to the locally observed road surface plane, and the leveling controller maintains the sprung vehicle body 4 in its calibrated attitude relative to the road. In the gravity-relative static mode, the leveling controller compares the road surface plane orientation based on lane lines with the orientation based on the horizon to separate the estimation of the road longitudinal and transverse slopes from the attitude of the sprung vehicle body relative to the road. The leveling controller only allows the issuance of leveling commands when quasi-static conditions are met. Quasi-static conditions include: vehicle speed below a threshold; parking brake or driving brake. The motion has been applied; or the vehicle 1 is determined to be in a quasi-static state based on at least one of longitudinal acceleration, lateral acceleration, yaw rate, steering angle, wheel speed change, and image motion. During driving, the leveling controller uses low-pass filtering, time window, and dynamic compensation, and excludes transient pitch during braking and transient roll during turning from static height correction. When the vehicle 1 is driving, the pitch and roll values ​​determined by the image processing module are used as closed-loop feedback quantities, and the height value obtained by the vehicle height sensor 6 is used as auxiliary feedback quantities. When the vehicle 1 is stationary or at low speed, the height value is used as the main feedback quantity, and attitude correction is performed based on the pitch and roll values.

[0104] Specifically, this invention sets up a road relative mode, a gravity relative static mode, quasi-static judgment conditions, and different feedback quantity allocation strategies for driving and stationary conditions. By comparing the direction separation of road slope and vehicle body posture based on lane lines and horizon, it can distinguish between road longitudinal slope, cross slope and vehicle body posture deviation, avoid mistaking road slope for suspension error, and prevent incorrect leveling under dynamic conditions such as braking and turning, thereby improving the accuracy, safety and adaptability of leveling control.

[0105] In this embodiment, the forward-facing camera 5 is a single forward-facing camera, which is fixed to the inside of the windshield or the front of the roof of the sprung vehicle body 4, and has intrinsic and extrinsic parameters calibrated relative to the vehicle coordinate system; the vehicle height sensor 6 includes a linkage sensor; the pitch and roll values ​​determined by the image processing module can be used for the leveling control of the leveling controller, and can also be used to detect camera calibration drift caused by changes in suspension or load; the leveling controller includes a deviation determination unit, a control quantity generation unit, and a drive unit; the deviation determination unit determines the deviation between the current vehicle posture and the target state based on the height, pitch, and roll values; the control quantity generation unit generates the target height, target pressure, or target gas quantity of each air suspension element 3 based on the deviation; the drive unit controls the inflation, deflation, or pressure holding of the air suspension elements 3 based on the target height, target pressure, or target gas quantity.

[0106] Specifically, the visual attitude information of this invention serves both leveling control and ADAS camera calibration drift detection. This means it can reuse the existing ADAS camera and vehicle network of vehicle 1 to achieve closed-loop air suspension leveling and promptly detect camera calibration drift caused by changes in suspension or load, improving the collaborative reliability of the vehicle's perception system and suspension system. The forward-facing camera 5 is an existing ADAS forward-facing camera of vehicle 1, calibrated with intrinsic and extrinsic parameters. Its image data is transmitted to the image processing module via the vehicle network bus. The system does not have a dedicated suspension attitude camera.

[0107] Furthermore, the leveling controller is configured with different driving heights, different loading heights, bus kneeling descent, trailer connection, docking, ferry, maintenance, or static leveling states. This allows the system to adapt to the vehicle's posture requirements in different usage scenarios, improving the functionality and practicality of the air suspension leveling system.

[0108] As an optimization, the vehicle height sensor 6 includes a main sensor and at least one redundant sensor. The vehicle height sensor 6 is located at the center of the axle or on one side. It can improve the reliability of height measurement while maintaining the reduction of the number of sensors, and support degraded operation or rationality verification when the main sensor is abnormal.

[0109] As an optimization, the vehicle height sensor 6 includes three vehicle height sensors 6, which are respectively arranged at the left front, right front and rear axle positions of the vehicle 1, and the number of vehicle height sensors 6 is less than the number of controllable air suspension elements 3; it can obtain the vehicle's height information in the front-rear and left-right directions with fewer physical sensors, and cooperate with the visual pitch and roll values ​​to achieve reliable vehicle attitude estimation and leveling control.

[0110] As an optimization, the air suspension element 3 includes an air spring, a solenoid valve connected to the air spring, a compressor, an air tank, and an exhaust valve. The leveling controller can precisely adjust the amount of gas in each air spring through inflation, deflation, and pressure holding, thereby achieving closed-loop control of vehicle height, pitch, and roll.

[0111] Specific working principle: When this system is working, the forward-facing camera 5 acquires real-time images of the road ahead and sends them to the image processing module. The visual attitude estimator selects one or more methods based on the current scene, including lane-line based, horizon-based, and backup methods when there are no lane lines or no horizon. It extracts features such as lane convergence, vanishing point, horizon tilt, and edge direction statistics to estimate the pitch and roll values ​​of the sprung vehicle body 4 relative to the reference attitude and outputs the corresponding confidence scores. At the same time, the leveling controller reads the absolute height value provided by the vehicle height sensor 6 and uses this height value, along with the visual pitch and roll values, to constitute the current leveling state. The system compares the values ​​with the preset target state, and the deviation determination unit calculates the height, pitch, and roll deviations. The control quantity generation unit then generates the target height, target pressure, or target gas quantity for each air suspension element 3. The drive unit controls the solenoid valve, compressor, and exhaust valve to inflate, vent, or maintain the pressure of the corresponding air springs. After the air suspension element 3 is activated, the forward-facing camera 5 captures images again, the image processing module re-estimates the pitch and roll values, and the leveling controller reads the height value again and updates the leveling status, forming a visual closed-loop feedback. The above process is repeated until the height, pitch, and roll errors enter the allowable tolerance range.

[0112] During the adjustment process, if the confidence level of the visual attitude estimator output is lower than the threshold, or if the vehicle dynamic parameters indicate that the quasi-static conditions are not currently met, the leveling controller suspends visual closed-loop leveling to maintain the final safe state of each valve in the air suspension component 3, or enters a degraded function mode based on the remaining vehicle height sensor 6. During the transition period when the confidence level crosses the threshold, the fusion module applies a rate limit to the fusion gain of the visual observations to suppress leveling state transitions.

[0113] In addition, when the vehicle 1 is stationary or in a low-dynamic state, the leveling controller can apply short-term inflation or deflation excitation to selected air suspension components 3. Utilizing the pitch and roll angle differential responses measured by the forward-facing camera 5 and the vertical response measured by the vehicle height sensor 6, it identifies or updates the actuator geometry matrix online, compensates for load-related aerodynamic characteristics, detects air springs or valves with insufficient performance, and verifies whether the sign and amplitude of the visual attitude signal meet expectations. Simultaneously, the leveling controller can be configured with a road-relative mode and a gravity-relative stationary mode. By comparing the road surface plane direction based on lane lines with the direction based on the horizon, it separates and estimates the road longitudinal and transverse slopes from the vehicle's attitude relative to the road, avoiding misclassifying road slope as suspension error. The image processing module outputs the road surface plane normal based on lane lines and the gravity reference direction based on the horizon in parallel. When the deviation exceeds the reasonable slope range, the leveling controller determines it as road slope and maintains the calibrated attitude relative to the road in road-relative mode. When the deviation is within the suspension error range, it determines it as vehicle attitude deviation and performs closed-loop correction.

[0114] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A vehicle air suspension leveling system based on a small number of height sensors and a visual closed loop, comprising a vehicle (1), said vehicle (1) comprising a vehicle body (2) and a sprung body (4) supported on the vehicle body (2) by a plurality of controllable air suspension elements (3), characterized in that, The vehicle air suspension leveling system includes: A forward-facing camera (5) is fixed to the sprung body (4) and calibrated relative to the vehicle coordinate system, and is capable of collecting forward image data. The vehicle height sensor (6) is located between the vehicle body (2) and the sprung body (4), and its number is less than that of the air suspension components (3). It can detect the distance between the vehicle body (2) and the sprung body (4). The image processing module is able to determine the pitch and roll values ​​of the sprung body (4) relative to the reference attitude based on the image data collected by the forward-facing camera. The leveling controller can control the air suspension element (3) based on the height value obtained by the vehicle height sensor (6) and the leveling state represented by the pitch and roll values ​​determined by the image processing module, so that the vehicle body posture reaches the target state. When the vehicle (1) is stationary or in a low dynamic state, the leveling controller applies a short-term inflation or deflation command to the selected air suspension element (3). Based on the pitch angle and roll angle differential response measured by the forward camera (5) and the vertical response measured by the vehicle height sensor (6), the controller identifies or updates the actuator geometry matrix, detects the control valve of the air suspension element (3) with insufficient performance, and verifies whether the sign and amplitude of the visual posture signal meet the expectations. The image processing module outputs the road surface plane normal based on the lane lines and the gravity reference direction based on the horizon in parallel. When the deviation between the two exceeds the reasonable slope range, the leveling controller determines it as road slope and maintains the calibration attitude relative to the road in the road relative mode. When the deviation is within the suspension error range, it determines it as vehicle body attitude deviation and performs closed-loop correction. The forward-facing camera (5) is an existing ADAS forward-facing camera on the vehicle (1) that has been calibrated with internal and external parameters. Its image data is transmitted to the image processing module via the vehicle network bus. The system does not have a dedicated suspension attitude camera.

2. The vehicle air suspension leveling system based on a small number of height sensors and visual closed-loop as described in claim 1, characterized in that, The image processing module includes a visual pose estimator configured to select from a variety of complementary pose estimation methods to determine pitch and roll values. The pose estimation methods include: The lane-line-based method obtains at least one of longitudinal vanishing point, apparent convergence, perspective scale, and left-right asymmetry from the lane lines in the forward image data, and determines the pitch and / or roll values ​​accordingly. The horizon-based method detects the horizon in the foreground image data, determines the yaw value based on the tilt angle of the horizon, and determines the pitch value based on the calibrated vertical image position. And an alternative method for situations where there are no lane lines and / or no horizon, which obtains the roll value through edge detection and robust statistics on the edges of the near-horizontal and vertical scene, and obtains one or more vanishing points through the intersections or convergences of selected families of lines, so as to estimate the pitch value based on the one or more vanishing points; The leveling controller determines whether the visual feedback is effective based on the confidence level output by the visual attitude estimator, and controls the air suspension components (3) based on the pitch and roll values ​​determined by the image processing module when the visual feedback is effective.

3. The vehicle air suspension leveling system based on a small number of height sensors and visual closed-loop as described in claim 2, characterized in that, The lane line-based method includes: when a longitudinal lane line or road boundary is detected, fitting left and right lane models and determining their convergence in the image; The roll value is determined based on the lane geometric rotation caused by roll changes and the asymmetric changes in the apparent lane width, slope, or perspective scale on both sides of the image. The pitch value is determined based on the vertical displacement of the longitudinal vanishing point caused by pitch changes and the apparent convergence and scale change patterns of lane lines. The measured features are compared with the features predicted based on camera calibration and vehicle nominal attitude. Furthermore, lane measurements are accumulated across multiple image frames and robustly fitted using RANSAC, M-estimation, or Kalman filtering. For curved lanes, the lane model is represented by a polynomial or spiral and local extrapolation is performed to obtain the vanishing direction; based on the road model and uncertainty estimation, it is not required that the boundaries of the two lanes be completely parallel in the image.

4. The vehicle air suspension leveling system based on a small number of height sensors and visual closed-loop as described in claim 2, characterized in that, The horizon-based method includes: identifying horizons formed by skylines, road and sky boundaries, distant terrain boundaries, and building lines through a horizon detector, or inferring horizons statistically based on horizontal texture and motion features; The tilt value is determined by the inclination of the horizon relative to the x-axis of the image, and the pitch value is determined by its vertical position in combination with the camera's intrinsic parameters and nominal calibration. When the horizon is obscured, is significantly nonlinear, or inconsistent with other visual features, the corresponding estimate is rejected.

5. The vehicle air suspension leveling system based on a small number of height sensors and visual closed-loop as described in claim 2, characterized in that, The alternative methods for situations where there are no lane lines and / or no horizon include: performing edge detection, line segment detection, or learned structural feature extraction; Line segments are selected based on their length, contrast, duration, semantic category, and proximity to the horizontal or vertical direction after compensation based on the current pose assumption; an edge direction histogram or probability distribution is formed, and the dominant rotation of the near-horizontal and near-vertical line families is used as the tilt value. Based on the vertical image position of the vanishing point, the convergence or divergence of near-vertical scene edges, the perspective scale of repeating structures, or the pitch value estimated from the road surface plane obtained by optical flow or monocular depth, multiple vanishing points are used in the structured scene geometry model to exclude dynamic objects through semantic segmentation, motion inconsistency, or multi-frame tracking.

6. The vehicle air suspension leveling system based on a small number of height sensors and visual closed-loop as described in claim 2, characterized in that, The leveling controller determines the effectiveness of visual feedback based on confidence level, including providing an estimate and confidence level for each visual method, wherein the confidence level depends on at least one of the following: number and distribution of features, fitting residuals, inter-frame consistency, image exposure, visibility, weather, camera occlusion, road curvature, and consistency between independent methods. The image processing module also includes a fusion module, which effectively observes by combining weighted averaging, filtering, Bayesian estimation, or learning models. The fusion module continuously modulates the fusion gain of visual observations in the leveling state estimation with confidence. The lower the confidence, the higher the weight of the physical height channel. During the transition period when the confidence crosses the threshold, a rate limit is applied to the fusion gain to suppress the leveling state jump. When the confidence is lower than the threshold, visual closed-loop leveling is paused, and the system maintains the last safe state of each valve of the air suspension element (3), or enters the degraded function mode based on the remaining vehicle height sensor (6).

7. The vehicle air suspension leveling system based on a small number of height sensors and visual closed-loop as described in claim 1, characterized in that, The vehicle body height sensor (6) measures a height z s at a known vehicle body plan coordinate (x s , y s ) as a vertical translation reference, and calculates a virtual vehicle body height z i for a suspension position coordinate (x i , y i ) based on a pitch value θ and a roll value φ determined by the image processing module. z i = z s + (x i - x s ) tan(θ) + (y i - y s ) tan(φ) and the calculated virtual vehicle body height z i as a virtual corner point height signal of the levelling controller; There are three vehicle height sensors (6), which are respectively arranged at the left front, right front and rear axle positions of the vehicle (1); the leveling controller will verify the consistency between the pitch / roll angle reconstructed by the height difference of any two vehicle height sensors (6) and the pitch / roll value output by the image processing module. When the deviation continues to exceed the limit, the camera external parameter calibration will be updated to compensate for calibration drift. The physical height value provided by the vehicle height sensor (6) is used to provide vertical translation information, and the pitch and roll values ​​provided by the image processing module are used to provide the two rotational degrees of freedom required to determine the vehicle plane.

8. The vehicle air suspension leveling system based on a small number of height sensors and visual closed-loop as described in claim 1, characterized in that, The leveling controller is equipped with a road-relative mode and a gravity-relative stationary mode; In the road relative mode, the image processing module determines the orientation of the forward-facing camera (5) relative to the locally observed road surface plane, and the leveling controller keeps the sprung vehicle body (4) in its calibrated attitude relative to the road. In the gravity-relative static mode, the leveling controller compares the road surface plane orientation based on lane lines with the orientation based on the horizon to separate the road longitudinal and transverse slopes from the attitude of the sprung vehicle body relative to the road. The leveling controller is only allowed to issue leveling commands when quasi-static conditions are met, including: vehicle speed is below a threshold; parking brake or service brake has been applied. Alternatively, the vehicle can be judged to be in a quasi-static state based on at least one of the following: longitudinal acceleration, lateral acceleration, yaw rate, steering angle, wheel speed change, and image motion.

9. The vehicle air suspension leveling system based on a small number of height sensors and visual closed-loop as described in claim 1, characterized in that, During driving, the leveling controller employs low-pass filtering, time window, and dynamic compensation, and eliminates transient pitch during braking and transient roll during turning from static height correction; When the vehicle (1) is moving, the pitch and roll values ​​determined by the image processing module are used as closed-loop feedback quantities, and the height value obtained by the vehicle height sensor (6) is used as auxiliary feedback quantities; when the vehicle is stationary or at low speed, the height value is used as the main feedback quantity, and attitude correction is performed based on the pitch and roll values.

10. The vehicle air suspension leveling system based on a small number of height sensors and a visual closed loop according to any one of claims 1-9, characterized in that, The forward-facing camera (5) is one or more of a monocular camera, a stereo camera, a fisheye camera, or a multi-camera group. The forward-facing camera (5) is fixed to the inside of the windshield or the front of the roof of the sprung vehicle body (4) and has internal and external parameters calibrated relative to the vehicle coordinate system. The vehicle height sensor (6) includes a linkage sensor, an axle position sensor, or a pressure / load fusion model verified by a physical sensor; The pitch and roll values ​​determined by the image processing module can be used for the leveling control of the leveling controller, and can also be used to detect camera calibration drift caused by changes in suspension or load. The leveling controller includes a deviation determination unit, a control quantity generation unit, and a drive unit. The deviation determination unit determines the deviation between the current vehicle body posture and the target state based on the height value, pitch value, and roll value. The control quantity generation unit generates the target height, target pressure, or target gas quantity for each air suspension element based on the deviation. The drive unit controls the inflation, deflation, or pressure maintenance of the air suspension elements based on the target height, target pressure, or target gas quantity.