Method and system for evaluating and predicting top passing safety of unmanned vehicle

Abstract

The invention discloses an evaluation and prediction method and system for the top passing safety of an unmanned vehicle, and aims to solve the technical defect that the traditional passing ability evaluation neglects the risk of a top space. The height of a top obstacle and the three-dimensional contour information of a road surface are accurately collected through a multi-sensor fusion CFI technology of a laser radar and a camera; calculating a dynamic height increment in combination with vehicle static parameters to obtain an actual maximum height; safety redundancy is introduced to judge trafficability, and passing, braking or lane changing instructions are output; according to the invention, trafficability evaluation is expanded from a two-dimensional ground to a three-dimensional space, top risk look-ahead pre-judgment is realized, the traffic safety and efficiency in a complex scene are improved, and the method is compatible with an existing automatic driving system.

Description

Technical Field

[0001] This invention relates to the field of autonomous vehicle control, and more particularly to a method and system for assessing and predicting the safety of overpasses on the roof of autonomous vehicles. Background Technology

[0002] Vehicle passability is one of the core evaluation indicators in the field of vehicle engineering. It is defined as the ability of a vehicle, under rated load and at a sufficiently high average speed, to traverse various rough roads, off-road areas, and overcome various obstacles. The assessment of this capability heavily relies on a series of static geometric and mechanical parameters, including but not limited to ground clearance, approach angle, departure angle, breakover angle, turning radius, and coefficient of friction. This mature evaluation system primarily addresses the problem of vehicle passability on two-dimensional terrain, ensuring that the vehicle's undercarriage does not interfere with ground obstacles ("no bottoming out") and that it possesses continuous traction, thus forming the theoretical basis for off-road vehicle design.

[0003] However, with the rapid development of autonomous driving technology, the limitations of the aforementioned traditional theories have become increasingly apparent. Their fundamental flaw lies in the fact that traditional obstacle clearance is a purely two-dimensional concept, completely ignoring the risks posed by overhead obstacles to vehicles in three-dimensional space. In human driving scenarios, drivers can subjectively judge and compensate for overhead space through visual observation (such as height restriction bars); however, in autonomous driving scenarios, the unpredictability of the vehicle's overhead space—such as low tunnel domes, crossbeams in multi-level parking garages, hanging tree branches, or temporary falling objects—becomes a technical blind spot restricting the safe and automated operation of vehicles across all areas and conditions. Traditional theory treats the vehicle roof as a static, passive protective shell, which can no longer meet the urgent needs of intelligent connected vehicles for comprehensive perception and interaction with complex environments.

[0004] At the same time, although the traditional vehicle passability theory is quite mature, its limitations are becoming increasingly apparent in the face of the development needs of autonomous vehicles, mainly in the following aspects: 1. The theoretical dimension is lacking, and there is a lack of a definition and evaluation system for the passability of the three-dimensional top space.

[0005] Existing technologies strictly limit "roof passability" to the interaction between the vehicle and two-dimensional terrain, with the core being to prevent the chassis from interfering with ground obstacles (i.e., "not bottoming out"). This theory completely ignores the risk of overhead obstacles that vehicles face in the vertical direction and fails to establish any definition, evaluation index, or predictive model for "roof passability," resulting in a theoretical gap in the assessment of autonomous vehicles' passability in three-dimensional space.

[0006] 2. Insufficient perception capabilities; the environmental perception system has a "blind spot at the top".

[0007] Current perception systems for autonomous vehicles (such as front-facing cameras and forward-facing radar) are primarily optimized for traffic scenarios in the horizontal and pitch directions. Their sensor layout and perception algorithms do not consider obstacles directly above the vehicle's roof as core detection targets. For complex overhead obstacles such as low tunnel walls, beams in multi-level parking garages, and drooping tree branches, existing systems struggle to achieve stable and accurate detection and ranging, failing to provide reliable overhead environmental information for decision-making and control.

[0008] 3. The decision-making logic is missing, and the planning and control algorithm does not consider the top gap.

[0009] Based on existing ground clearance theories, the path planning and motion control modules of autonomous vehicles rely solely on ground clearance (such as slopes and ditches) and traffic rules for decision-making, completely neglecting dynamic "top clearance." This can lead to vehicles planning or executing dangerous paths that, while having unobstructed ground, could result in collisions between the vehicle's roof and obstacles. For example, when entering a bridge with insufficient height restrictions, the vehicle might fail to anticipate the situation and drive straight in, causing an accident.

[0010] 4. Limited functionality expansion, unable to support roof-based active interactive applications.

[0011] With the development of autonomous driving technology, vehicle roofs are evolving into key interactive platforms for automatic charging, drone logistics connections, and modular functional expansion. Existing technologies are unable to dynamically predict and assess the relative position, attitude, and safety between the vehicle roof and external devices (such as charging arms and drones), which severely restricts the realization and reliability of these new top-level interactive functions. Summary of the Invention

[0012] To address the shortcomings of existing technologies, this invention provides a method for assessing and predicting the top-passing safety of unmanned vehicles. By introducing the impact of dynamic changes in vehicle attitude, such as rear-end lifting when driving into a pothole or front-end lifting when running over an obstacle, on the actual passing height, a precise calculation model of "static vehicle height + dynamic increment" is established. This model can identify potential collision risks caused by changes in vehicle attitude in advance, thereby making decisions such as slowing down or detouring, fundamentally avoiding the safety hazards caused by traditional methods that rely solely on static vehicle height for judgment.

[0013] To achieve the above objectives, the present invention provides the following technical solution: The safety assessment and prediction methods for the top of autonomous vehicles include the following steps: a. Construct a theoretical model of vehicle roof passability, which includes vehicle roof passability criteria and dynamic height calculation methods under flat road surfaces, sloping road surfaces, pothole road surfaces, and road surfaces with raised obstacles; b. Deploy a multi-sensor perception system consisting of LiDAR and cameras, and collect information on obstacles above the vehicle in front of it and three-dimensional contour information of the road surface through multi-sensor fusion technology; c. Based on the theoretical model and the collected information, calculate the actual maximum height of the vehicle during its driving process in real time; d. Compare the vehicle's actual maximum height with the effective height of the obstacle above, and combine this with a preset safety redundancy to generate a top clearance assessment result; Based on the evaluation results, e outputs corresponding control commands to control the vehicle to perform operations such as passing, braking to stop, or changing lanes to bypass.

[0014] The present invention further defines the technical solution as follows: Preferably, the passability criterion under the smooth road surface is as follows: let the vertical height of the highest point of the vehicle be Hv, and the lowest point of the effective height of the top obstacle be Hz. When Hv < Hz, the vehicle can pass safely; when Hv ≥ Hz, the vehicle cannot pass. The actual maximum height H of vehicles under the inclined road surface and the pothole road surface all =asinθ+H(cosθ-1)+Hv, where a is the horizontal distance from any point at the rear of the vehicle to the point where the rear wheel touches the ground, H is the initial height of that point, θ is the rotation angle of the vehicle body around the point where the rear wheel touches the ground, and Hv is the static highest point height of the vehicle; The actual maximum height H of the vehicle on the road surface with protruding obstacles. all =Hv+Δh1, where Δh1=x・d1 / L, x is the horizontal distance from the target point to the rear wheel, d1 is the height of the obstacle, and L is the wheelbase of the vehicle.

[0015] Preferably, the multi-sensor fusion technology uses the CFI formula to achieve data fusion, and the CFI formula is:

[0016] Where IOU is the intersection-union ratio of the camera detection point set S and the lidar detection point set J, i.e., IOU = (S∩J) / (S∪J), d is the normalized distance between the center points of the two detection boxes, sim is the cross-modal feature similarity, Cs is the camera detection confidence, and CM is the lidar detection confidence. The aforementioned camera detection point set S includes two-dimensional bounding boxes, segmentation information, visual features, and confidence levels; the lidar detection point set J includes 3D point clusters, 3D bounding boxes, reflection intensity / depth features, and confidence levels.

[0017] Preferably, the three-dimensional feature information of the top obstacle includes the obstacle's height above the ground, outline dimensions, and relative distance, which are obtained through the following methods: the lidar performs rasterization processing, cluster analysis, and outline extraction on the collected point cloud data, and calculates the height and position of the obstacle by combining geometric features; the camera obtains the regional parallax through binocular vision technology, converts it into relative distance after primary filtering to remove outliers, and then performs secondary filtering based on the road surface model to obtain the obstacle's height above the ground; the detection results of the lidar and the camera are fused using the CFI formula of claim 3 to obtain the accurate value of the obstacle's height above the ground.

[0018] Preferably, the three-dimensional contour information of the road surface includes the depth of the pit, the height of the raised obstacle, and the road surface slope, which is collected in the following way: the lidar is installed at the front of the vehicle body with a preset pitch angle, and the reference distance l0 of the flat road surface and the measurement distance l1 of the obstacle surface are collected. The height of the road obstacle is calculated according to the formula H1=( l1−l0) cosφ, where φ is the installation pitch angle of the lidar. The camera calculates the height and width of the obstacle through geometric modeling based on preset camera parameters, including focal length f, installation height H, pixel size px / py, and image coordinates. The detection results of the lidar and the camera are fused using the CFI formula of claim 3 to obtain the precise values ​​of the road surface feature parameters.

[0019] Preferably, the preset safety redundancy is 0.1 meters, when the actual maximum height H of the vehicle is... all When <Hz-0.1m, output pass command; when H all When the speed is ≥Hz-0.1m, a braking stop or lane change command is output; this control command is executed through an execution module, which includes a motor controller, a braking control computer, and a steering control computer.

[0020] This invention provides a system for assessing and predicting the safety of top-passing maneuvers on unmanned vehicles, comprising: The perception module consists of a lidar and a binocular camera. The lidar is used to collect 3D point cloud data of the road surface and overhead obstacles, while the binocular camera is used to collect image information of the environment in front. The two work together to obtain environmental perception data. The decision-making module includes a passability control computer and a data storage unit. The data storage unit pre-stores vehicle static parameters and preset safety redundancy. The passability control computer is used to load the vehicle roof passability theoretical model, receive and fuse environmental data from the perception module, execute the evaluation and prediction process described in claims 1, 3-6, and generate passability determination results and control commands. The execution module, including the motor controller, braking system controller and steering system controller, is used to receive control commands output by the decision module and drive the vehicle to complete passing, braking or lane changing operations.

[0021] The aforementioned vehicle static parameters include vehicle length, wheelbase, and the static maximum height Hv of the suspension model.

[0022] Beneficial effects Compared with existing technologies, it has the following advantages: This invention offers high safety and avoids collision risks. By introducing the impact of dynamic vehicle posture changes on clearance height, it establishes a "static vehicle height + dynamic increment" calculation model to accurately quantify the amount of vehicle body lift under road conditions such as potholes and bumps. This allows for the early identification of potential collision risks caused by changes in vehicle posture, enabling decisions such as slowing down or detouring, thus fundamentally avoiding the safety hazards caused by relying solely on static vehicle height for judgment. This invention provides forward-looking prediction to improve traffic efficiency. By fusing perception data from LiDAR and cameras, it can obtain three-dimensional contour information of the road ahead in advance and predict the attitude changes and final actual height of vehicles as they pass. This forward-looking capability allows vehicles to plan a smooth and efficient passage strategy in advance, rather than taking emergency braking when a risk is imminent, thus ensuring driving efficiency and comfort.

[0023] 3. This invention has high reliability and can perceive complementary results. By adopting a fusion perception scheme of lidar and camera, it combines the advantages of lidar's accurate ranging and camera's rich texture information. The two complement each other, effectively improving the recognition accuracy and reliability of height restriction pole height and road features under complex lighting and adverse weather conditions, providing a solid data foundation for safety assessment.

[0024] This invention solves key problems that existing technologies cannot address, such as top collision risk, unreliable detection in complex scenarios, and limited functional expansion, through a combination of three-dimensional theory, fusion perception, dynamic decision-making, and closed-loop control. At the same time, it introduces a safety redundancy design and combines it with changes in the vehicle's dynamic attitude to make the passability judgment more in line with real road conditions, avoid the rigidity defects of traditional static threshold judgment, and improve the robustness of the system. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the structure of the unmanned vehicle top-pass safety prediction system of the present invention. Figure 2 This is a schematic diagram illustrating the vehicle roof's passability under inclined and potholed road surfaces in Embodiment 1. Figure 3 This is a schematic diagram of the road surface information detected by lidar in Embodiment 1. Figure 4This is the camera-based road obstacle detection model in Example 1. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention. Example

[0027] This embodiment provides a system for assessing and predicting the safety of top-passage maneuvers for unmanned vehicles, such as... Figure 1 As shown, it includes: The perception module consists of a lidar 2 and a binocular camera 1. The lidar is used to collect three-dimensional point cloud data of the road surface and overhead obstacles, while the binocular camera is used to collect image information of the environment in front. The two work together to obtain environmental perception data. The decision-making module includes a passability control computer 3 and a data storage unit. The data storage unit pre-stores vehicle static parameters (vehicle length, wheelbase, height of the highest point of the suspension model ji static height Hv) and preset safety redundancy. The passability control computer is used to load the vehicle roof passability theoretical model, receive and integrate environmental data from the perception module, execute the evaluation and prediction process, and generate passability judgment results and control commands. The execution module, including the motor controller 4, the braking system controller 5, and the steering system controller 6, is used to receive control commands output by the decision module and drive the vehicle to complete passing, braking, or lane changing operations.

[0028] Based on the above system, this embodiment provides a method for assessing and predicting the safety of top-passing manned vehicles, including the following steps: a. Construct a theoretical model of vehicle roof passability, which includes vehicle roof passability criteria and dynamic height calculation methods under flat road surfaces, sloping road surfaces, pothole road surfaces, and road surfaces with raised obstacles; b. Deploy a multi-sensor perception system consisting of LiDAR and cameras, and collect information on obstacles above the vehicle in front of it and three-dimensional contour information of the road surface through multi-sensor fusion technology; c. Based on the theoretical model and the collected information, calculate the actual maximum height of the vehicle during its driving process in real time; d. Compare the vehicle's actual maximum height with the effective height of the obstacle above, and combine this with a preset safety redundancy to generate a top clearance assessment result; Based on the evaluation results, e outputs corresponding control commands to control the vehicle to perform operations such as passing, braking to stop, or changing lanes to bypass.

[0029] The theoretical basis for the vehicle roof passing through the height restriction barrier on the above-mentioned smooth road surface is as follows: Under ideal working conditions with a smooth road surface, the passability criterion between the vehicle and the height restriction barrier can be simplified into a purely static geometric interference problem, the core theoretical basis of which is as follows: As a rigid body, the vehicle's outline is considered a fixed envelope in three-dimensional space, while the height restriction bar is considered a spatial constraint plane with a definite height in the vertical direction. When a vehicle attempts to pass over the height restriction bar along a horizontal road surface, the passability criterion depends on the highest point on the vehicle's envelope. H v The lowest point H of the constraint plane represented by the height restriction bar z The vertical spatial relationship between them This criterion can be expressed as: Conditions under which the test will fail: .

[0030] If the vertical height of the vehicle's highest point (relative to a flat road surface) is greater than or equal to the lowest point of the effective height of the height restriction bar, then the vehicle's envelope plane and the constraint plane of the height restriction bar will geometrically intersect or be tangent. This indicates that a direct structural interference (i.e., collision) will occur between the vehicle and the height restriction bar, and the vehicle will not be able to pass safely.

[0031] Conditions under which it can be passed: .

[0032] If the vertical height of the highest point of the vehicle is strictly less than the lowest point of the effective height of the height restriction bar, then the vehicle's envelope is completely below the constraint plane of the height restriction bar. There is a positive safety clearance between the two in the vertical direction. This clearance ensures that the vehicle will not theoretically come into contact with the height restriction bar, and therefore can pass safely.

[0033] The theoretical basis for vehicles passing over height restrictions on sloping roads and potholes: like Figure 2 As shown, when the front wheels of a car drive into a pit or slope of depth d, while the rear wheels remain on level ground, the car body will rotate clockwise around the rear wheel contact point O (viewed from the right side of the vehicle). To quantify the resulting rise in the rear of the car, the following geometric model is established: Let the wheelbase of the vehicle be L, and let any point P at the rear of the car be located at a horizontal distance a and an initial height H behind the rear wheel contact point O. The X-axis is positive, pointing horizontally towards the front of the car, and the Y-axis is positive, pointing vertically upward.

[0034] The coordinates of the front wheel center are (L, R), where R is the wheel radius, and the coordinates of point P are (-a, H). When the front wheel descends by d, the coordinates of the front wheel center become (L, Rd). The car body rotates clockwise around O by an angle θ (viewed from the right side). θ > 0 indicates that the rear of the car rises, based on geometric relationships: ; Around the origin O Turn clockwise θ Transformation formula (from initial position to...) θ ), ; The initial coordinates of P are X=-a, Y=H After rotation, Let the initial Y-coordinate be Y0=H.

[0035] The vehicle body height is: ; The distance between the highest point of any point behind the rear wheel of the raised car and the ground is: h all =Δh+H v =asinθ+H(cosθ-1)+ H v Therefore, the vehicle's ability to pass over uneven roads depends not only on the vehicle's height but also on the wheelbase and the depth of the potholes. Since the vehicle height and wheelbase are fixed, the vehicle's ability to pass over uneven roads is related to the depth of the potholes.

[0036] Theoretical basis for vehicle roof height restriction due to road surface protrusions After a car's front wheels pass over a raised obstacle, the height at which any point in front of the car rises depends primarily on factors such as the obstacle's height, the car's wheelbase, and the horizontal distance from the measuring point to the rear wheels. Let the height of the obstacle be... d 1. Wheelbase is L The horizontal distance from the target point to the rear wheel is x The height the target point rises is directly proportional to the horizontal distance from that point to the rear wheel. The height rise is directly proportional to the obstacle height. The height rise is inversely proportional to the wheelbase. It is independent of the obstacle's slope because once the front wheels reach the highest point of the obstacle, the angle of rotation of the car around the rear wheels depends only on the obstacle height and wheelbase; the slope does not affect the final height rise (assuming the car can successfully pass the obstacle).

[0037] Therefore, the equation for the height increase is: ; Δ h 1 represents the height raised to any point in front of the car, where the wheelbase is... L Assuming the target point is a fixed value, the vehicle's height depends only on the height of the obstacle. d 1. Related to this.

[0038] At this point, we have calculated the actual height the vehicle rises under various road conditions, including flat roads, downhill slopes, potholes, and raised obstacles, thus determining the vehicle's real-time height during travel. Based on this, by accurately detecting the height of overhead obstacles (such as height restriction barriers or bridge structures) using sensors and comparing it to the vehicle's current actual height, we can determine whether the vehicle can safely pass. Therefore, the next step will focus on utilizing sensors to achieve accurate measurement of the height of height restriction barriers.

[0039] superior Narrative Multi-sensor fusion technology uses the CFI formula to achieve data fusion. The CFI formula is: ; Where IOU is the intersection-union ratio of the camera detection point set S and the lidar detection point set J, i.e., IOU = (S∩J) / (S∪J), d is the normalized distance between the center points of the two detection boxes, sim is the cross-modal feature similarity, Cs is the camera detection confidence, and CM is the lidar detection confidence. The aforementioned camera detection point set S includes two-dimensional bounding boxes, segmentation information, visual features, and confidence levels; the lidar detection point set J includes 3D point clusters, 3D bounding boxes, reflection intensity / depth features, and confidence levels.

[0040] The three-dimensional feature information of the aforementioned top obstacle includes the obstacle's height above the ground, outline dimensions, and relative distance, which are obtained through the following methods: The LiDAR performs rasterization, cluster analysis, and outline extraction on the collected point cloud data, and calculates the obstacle's height and position based on geometric features; the camera acquires regional parallax using binocular vision technology, converts it into relative distance after primary filtering to remove outliers, and then performs secondary filtering based on the road surface model to obtain the obstacle's height above the ground; the detection results from the LiDAR and camera are fused using the CFI formula to obtain the precise value of the obstacle's height above the ground.

[0041] The aforementioned three-dimensional road surface contour information includes pothole depth, height of raised obstacles, and road surface slope, and is collected through the following methods: Figure 3 As shown, the lidar is mounted at an elevation angle φ on the front of the vehicle. Q Let Q1 and Q2 be the same spatial points that the vehicle passes through at different times. Q1 and Q2 form a horizontal line segment, and Q1Q2JM form a parallelogram. According to the properties of a parallelogram, we can get Q1M = Q2J. When encountering road obstacles, the distance between the lidar ranging point and the top surface of the obstacle will change abruptly. (Definition:) l0 = the baseline distance to the flat road surface; l1 = Measured distance to the top surface of the obstacle; (l0, l1) can be directly acquired by lidar. In the right triangle △JM1W, JM1 = Q2M1 − Q1M = l1 − l0. Therefore, the height of the road obstacle can be calculated using the following formula: ; The height H1 and width Q1Q2 of the obstacle can be calculated, and the relevant geometric information of the pit can also be calculated in the same way.

[0042] The camera captures road obstacles, such as... Figure 4 As shown, Given the camera height H, the distance between the world coordinate point corresponding to the image coordinate center and the camera on the Y-axis, the image coordinates of the lens center point, the image coordinates P1 of the measured pixel, and the pixel length p. y Pixel width p x Given a camera with focal length f and the highest point of the obstacle being Q, then: ; Based on the calculated distance between the obstacle and the camera, the result is obtained through simple addition and subtraction of data. EP The distance between them is the height of the obstacle. EQ The height is: ; Based on the above derivation process, the height and width of the obstacle can be obtained, enabling camera-based data acquisition and processing. By fusing the information collected by the lidar with the information collected by the camera using the CFI fusion method, a more accurate information about the obstacle height can be obtained.

[0043] The aforementioned preset safety redundancy is 0.1 meters. The vehicle can obtain the ground clearance (Hz) of the lower edge of the height restriction bar through sensors (a fusion system of lidar and camera). The actual clearance height of the vehicle needs to take into account dynamic attitude changes: when driving downhill or on a pothole, the rear suspension will rise when the front wheels enter, increasing the maximum vehicle height to H. v +Δh; When the front wheel rolls over a raised obstacle, the body on the front suspension axle side will rise, causing the maximum height to become H. v +Δh1, the final height of the vehicle can be expressed by the formula: ; When the actual maximum height H of the vehicle all When <Hz-0.1m, output pass command; when H all When l≥Hz-0.1 m, a braking stop or lane change command is output; this control command is executed through an execution module, which includes a motor controller, a braking control computer, and a steering control computer.

[0044] Example 2 Suppose an autonomous vehicle (hereinafter referred to as "the vehicle") is driving on an urban road with a static maximum height Hv of 2.2 meters. A height restriction bar appears ahead with a marked height of 2.5 meters. The vehicle's task is to safely assess whether it can pass through the height restriction bar without compromising driving smoothness and efficiency.

[0045] System Configuration The vehicle is equipped with LiDAR and cameras mounted on its roof. These two sensors are synchronized through hardware and integrated through software to form a joint perception system.

[0046] Computing unit, vehicle-mounted domain controller, or central computing platform.

[0047] The data source stores the vehicle's basic parameters, including vehicle length, wheelbase, suspension model, and static height Hv.

[0048] Implementation steps Step 1: Environmental Perception and Target Recognition The vehicle's fusion perception system continuously scans the road ahead. At approximately 50 meters from the height restriction barrier, the system performs the following tasks: The camera identifies the visual features of the height restriction pole using a visual algorithm and locates it initially. For precise distance and height measurement, lidar point cloud data was used to accurately calculate the absolute height (Hz) of the bottom edge of the height restriction pole from the ground. Through point cloud clustering and plane fitting, Hz was calculated to be 2.48 meters (slightly lower than the marked height). The road surface scanning and lidar simultaneously scanned the road surface in front of and behind the height restriction bar, detecting a small pothole caused by road maintenance at the position where the front wheels of the vehicle were about to pass. Step Two: Dynamic Height Calculation and Risk Assessment After receiving the sensing data, the computing unit starts the dynamic height prediction model; Based on the scenario analysis, the system determines that the front wheels will drive into a pothole. This scenario belongs to "front wheels passing through a downhill slope or pothole", and the corresponding vehicle posture is "rear lift". The system calculates the dynamic lift Δh based on parameters such as the depth and slope of the pit provided by the lidar, as well as the vehicle's wheelbase and center of gravity position, combined with the vehicle dynamics model. The system calculates in real time the expected lift height Δh of the rear overhang when the current wheel enters the pit. The calculated value is Δh = 0.08 meters. Calculate the actual maximum height H_actual: Applying the formula proposed in this patent: H_actual = Hv + Δh That is: H_actual = 2.20 + 0.08 = 2.28 meters; Step 3: Safety Decision and Vehicle Control The system compares the calculated actual maximum height H_actual with the sensed height of the height limit bar Hz in real time: H_actual (2.28 m) < Hz (2.48 m) Safety margin = Hz - H_actual = 0.20 m; Decision: The safety margin is sufficient (greater than the preset safety threshold of 0.1 m). The system determines that it is possible to pass. The vehicle control system will maintain the current vehicle speed or make minor speed adjustments to smoothly pass the height limit bar without unnecessary braking.

[0049] Comparative Example (Traditional Method as a Comparative Ratio): If the traditional method is adopted, only the static vehicle height Hv (2.20 m) and the height of the height limit bar Hz (2.48 m) are compared, and a safety margin of 0.28 m will be obtained. This method completely ignores the 0.08 m dynamic lift caused by the pothole. Although the traditional method also reaches the conclusion of "passable" in this example, its safety assessment is based on an incomplete model and there are potential risks. For example, if the pothole is deeper, resulting in Δh = 0.30 m, the traditional method will dangerously judge it as passable (2.20 < 2.48), while the method of this patent will accurately calculate H_actual = 2.50 m and trigger an alarm or braking to avoid collision (2.50 > 2.48).

[0050] Summary of the Embodiment: This embodiment clearly demonstrates how the method of the present invention realizes accurate and forward-looking assessment of the passing safety of the vehicle top through the integration of sensing, dynamic prediction and precise calculation. This method effectively compensates for the safety loopholes of the traditional static assessment model and ensures the passing safety and decision reliability of driverless vehicles under complex real road conditions.

[0051] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0052] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An evaluation and prediction method of top safety of an unmanned vehicle, characterized by, The method comprises the following steps: a. Constructing a roof passability theoretical model, which comprises vehicle roof passability criteria and dynamic height calculation methods under flat road surface, inclined road surface and pothole road surface, and road surface with protruding obstacles; b. Deploying a multi-sensor perception system composed of a laser radar and a camera, collecting vehicle front roof obstacle information and road surface three-dimensional contour information through multi-sensor fusion technology; c. Based on the theoretical model and the collected information, the actual maximum height of the vehicle during driving is calculated in real time; d. Comparing the actual maximum height of the vehicle with the effective height of the roof obstacle, and combining the preset safety redundancy, the roof passability evaluation result is generated; e. According to the evaluation result, the corresponding control instruction is output, and the vehicle is controlled to execute passing, braking and parking or lane changing operation.

2. The method of claim 1, wherein the method further comprises: The passability criterion under the flat road surface is: assuming that the vertical height of the highest point of the vehicle is Hv, and the lowest point of the effective height of the roof obstacle is Hz, when Hv<Hz, the vehicle can pass safely; when Hv≥Hz, the vehicle cannot pass; The actual maximum height H of the vehicle under the inclined road surface and the pothole road surface all =asinθ+H(cosθ-1)+Hv, wherein a is the horizontal distance from any point of the tail to the rear wheel contact point, H is the initial height of the point, θ is the rotation angle of the vehicle body around the rear wheel contact point, and Hv is the height of the static highest point of the vehicle. The actual maximum height Hv of the vehicle under the road surface with the convex obstacle all = Hv + Δhi, where Δhi = x - di / L, x is the horizontal distance from the target point to the rear wheel, di is the obstacle height, and L is the wheelbase of the vehicle.

3. The method of claim 1, wherein the method further comprises: The multi-sensor fusion technology realizes data fusion by using the CFI formula, and the CFI formula is: ; Wherein IOU is the intersection over union of the camera detection point set S and the laser radar detection point set J, that is, IOU=(S∩J) / (S∪J), d is the normalized distance between the centers of the two detection boxes, sim is the cross-modal feature similarity, Cs is the camera detection confidence, and CM is the laser radar detection confidence; The above camera detection point set S includes image two-dimensional boundary box, segmentation information, visual features and confidence; The laser radar detection point set J includes 3D point cluster, 3D boundary box, reflection intensity / depth feature and confidence.

4. The method of claim 3, wherein the method further comprises: The three-dimensional feature information of the roof obstacle includes the obstacle height, contour size and relative distance, which is obtained by: the laser radar performs rasterization processing, clustering analysis and contour extraction on the collected point cloud data, and calculates the height and position of the obstacle based on the geometric features; the camera obtains the regional parallax through binocular vision technology, converts it into relative distance after removing outliers through primary filtering, and then performs secondary filtering based on the road surface plane model to obtain the height of the obstacle from the ground; the detection results of the laser radar and the camera are fused through the CFI formula of claim 3 to obtain the accurate value of the obstacle height from the ground.

5. The method of claim 4, wherein the method further comprises: The road surface three-dimensional contour information includes pothole depth, protruding obstacle height and road surface slope, which is collected by: the laser radar is installed at the front of the vehicle body at a preset pitch angle, collects the reference distance l0 of the flat road surface and the measured distance l1 of the obstacle surface, and calculates the road obstacle height according to the formula H1=(l1-l0)cosφ, wherein φ is the installation pitch angle of the laser radar; The camera calculates the height and width of the obstacle based on the preset camera parameters focal length f, installation height H, pixel size px / py and image coordinates through geometric modeling; the detection results of the laser radar and the camera are fused through the CFI formula of claim 3 to obtain the accurate value of the road feature parameters.

6. The method of claim 1, wherein the method further comprises: The preset safety redundancy is 0.1 meters, when the actual maximum height H all When H all When H ≥ Hz-0.1 meters, output brake parking or lane bypass instruction; the control instruction is executed by an execution module, which includes a motor controller, a brake control computer and a steering control computer.

7. An autonomous vehicle roof-top passage safety assessment and prediction system, comprising: It comprises: The perception module consists of a lidar and a binocular camera. The lidar is used to collect 3D point cloud data of the road surface and overhead obstacles, while the binocular camera is used to collect image information of the environment in front. The two work together to obtain environmental perception data. The decision-making module includes a passability control computer and a data storage unit. The data storage unit pre-stores vehicle static parameters and preset safety redundancy. The passability control computer is used to load the vehicle roof passability theoretical model, receive and fuse environmental data from the perception module, execute the evaluation and prediction process described in claims 1, 3-6, and generate passability determination results and control commands. The execution module, including the motor controller, braking system controller and steering system controller, is used to receive control commands output by the decision module and drive the vehicle to complete passing, braking or lane changing operations. 8.The system according to claim 7, wherein, The vehicle's static parameters include vehicle length, wheelbase, and the static highest point height Hv of the suspension model.

Images