Adaptive empty parking space cleaning method based on point cloud obstacle perception

Abstract

The application discloses a kind of based on point cloud obstacle perception's adaptive empty parking stall cleaning method, this method includes: obtaining the local patrol scanning path of low-speed unmanned vehicle, vehicle pose, positioning state and obstacle point cloud data;Based on the local patrol scanning path and vehicle pose, build parking stall obstacle judgment area;The scale of obstacle point cloud in judgment area is counted to judge whether parking stall satisfies cleaning condition;The distance of front obstacle in the direction of vehicle cleaning is calculated to judge whether there is the risk of continuing to advance;Based on path effectiveness, obstacle point cloud scale, front obstacle distance and parking stall path release state, the parking stall cleaning state is dynamically updated by state machine, and adaptive cleaning control of empty parking stall is realized.The application is based on point cloud geometry processing and state machine decision, can effectively improve the safety, adaptability and operation efficiency of parking stall cleaning, and is suitable for low-speed unmanned environmental sanitation operation in complex scenes such as underground parking lot.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental perception and cleaning decision control for low-speed unmanned vehicles, and specifically relates to an adaptive empty parking space cleaning method based on point cloud obstacle perception. Background Art

[0002] With the continuous acceleration of urbanization, urban environmental sanitation management is facing the growing demand for refined operations. As an important and high-frequency operation link, parking space cleaning is widely distributed in large underground parking lots, commercial complexes, residential communities and public parking areas. The cleanliness of parking spaces not only affects the overall environmental sanitation level of the area, but also is directly related to vehicle access efficiency, parking safety and the vehicle usage experience of car owners. Especially during peak hours, clean empty parking spaces play a positive role in alleviating parking pressure and improving turnover efficiency.

[0003] At present, parking space cleaning operations mainly rely on two methods: one is traditional manual cleaning, in which sanitation workers carry tools or operate simple equipment to clean each space one by one. This method has high labor intensity, low efficiency, and the cleaning quality varies from person to person, making it difficult to ensure consistency and timeliness. The other is to use partially automated cleaning equipment, such as sweepers with preset trajectories or remote-controlled cleaning machines. Although such equipment can replace manpower to a certain extent, it is often only suitable for open, barrier-free continuous road surfaces, and has poor environmental adaptability to parking spaces with complex internal structures and dynamically changing space states, which often leads to missed cleaning, repeated cleaning or collision with obstacles, and the overall cleaning effect is not ideal.

[0004] In recent years, with the gradual maturity of unmanned driving technology, some enterprises have begun to introduce low-speed unmanned cleaning vehicles into the parking space cleaning scenario. This type of equipment relies on sensors such as LiDAR and cameras to achieve environmental perception and autonomous navigation, which improves the operation continuity and reduces the frequency of manual intervention compared with traditional methods. However, in the actual promotion, application and engineering implementation process, the existing unmanned cleaning solutions still expose the following significant deficiencies in the parking space scenario:

[0005] 1. The environmental perception dimension is single, and the real-time perception capability for obstacles inside parking spaces is insufficient. Most solutions plan cleaning trajectories based on global preset paths or fixed area coverage, and lack fine-grained detection and dynamic response capabilities for the actual obstacle distribution in parking spaces (such as adjacent vehicles, ground locks, cone barrels, and dropped items). Due to the narrow space of parking spaces and the random position of obstacles, it is difficult to accurately describe the real-time occupancy state in the parking space only by relying on the global map or visual features, resulting in the cleaning strategy cannot be flexibly adjusted with the obstacle distribution, which easily causes invalid cleaning, local missed cleaning or repeated operations.

[0006] 2. The dynamic target response mechanism during the cleaning process is weak, lacking effective distance judgment and state switching logic. During operation, moving or sudden targets such as pedestrians, vehicles entering, and temporary piles of objects may suddenly appear in front of the cleaning vehicle. If the system lacks accurate judgment of the distance to obstacles in front and an adaptive switching mechanism for the cleaning state based on this judgment, it faces a high risk of collision. At the same time, the cleaning process may be interrupted due to frequent avoidance, resulting in a significant decrease in overall operation efficiency.

[0007] 3. Some technical solutions rely excessively on visual recognition and deep learning models for obstacle classification and detection. While they perform well on specific datasets, real-world sanitation scenarios involve significant variations in lighting, blurred parking space boundaries, and diverse obstacle types, limiting the models' generalization ability. Furthermore, they place high demands on the computing power, storage space, and training data scale of embedded computing platforms, significantly increasing equipment hardware costs and system deployment complexity. This constraint is particularly pronounced in low-speed autonomous sanitation vehicles, limiting the large-scale application and ease of maintenance of such technologies.

[0008] 4. Existing solutions generally treat the cleaning process as a single, continuous task, failing to dynamically switch between stages and multiple strategies based on the actual parking space occupancy status, cleaning coverage progress, and vehicle positional deviations. When a portion of the parking space has been effectively cleaned, or when the vehicle's positional deviation occurs, the system still uses the initial global path, lacking the ability to replan the path and adjust the cleaning mode based on real-time perception data. This leads to a disconnect between cleaning execution and the actual on-site conditions, ultimately affecting the safety and integrity of the operation.

[0009] In summary, existing autonomous parking space cleaning solutions generally face several challenges in dynamic, confined, and frequently disruptive sanitation scenarios. These include coarse-grained real-time obstacle perception within the parking space, fixed and rigid cleaning paths, delayed response to sudden targets, and excessively high deployment costs associated with high-precision vision solutions. These shortcomings result in significant deficiencies in cleaning safety, environmental adaptability, and operational efficiency. Therefore, there is an urgent need for an empty parking space cleaning method that can comprehensively assess obstacle distribution within the parking space, distance to dynamic targets ahead, coverage of the sweeping path, and vehicle positional deviations based on low-cost point cloud perception, and adaptively adjust the cleaning stage and operational strategy accordingly. This would effectively improve the operational safety and cleaning efficiency of low-speed autonomous vehicles in such scenarios. Summary of the Invention

[0010] The present invention aims to at least solve one of the technical problems existing in the prior art. To this end, the present invention proposes an adaptive empty parking space cleaning method based on point cloud obstacle perception.

[0011] An adaptive empty parking space cleaning method based on point cloud obstacle perception according to an embodiment of the present invention includes the following steps:

[0012] S1. Acquire local patrol path, vehicle pose, positioning status, and point cloud data of obstacles around the vehicle for low-speed unmanned vehicles.

[0013] S2. Based on the local patrol path and the vehicle position, construct the parking space obstacle judgment area;

[0014] S3. Calculate the obstacle point cloud size within the obstacle judgment area of ​​the parking space, and determine whether the current parking space meets the cleaning conditions based on the obstacle point cloud size.

[0015] S4. Calculate the distance to obstacles ahead in the sweeping direction of the vehicle, and determine whether there is a risk of continuing to move forward during the sweeping process based on the distance to obstacles ahead;

[0016] S5. Based on path validity, obstacle point cloud size, distance to obstacles ahead, and parking space path publishing status, dynamically update parking space cleaning status to achieve adaptive cleaning control of empty parking spaces.

[0017] In some embodiments of the present invention, the step of obtaining the local scanning path in step S1 includes:

[0018] Read the global patrol path, obtain the serial number of the vehicle's current location in the global patrol path, and publish the local patrol path with a preset number of nearby points following the current serial number;

[0019] The local patrol path is a sequence of equally spaced path points.

[0020] In some embodiments of the present invention, the step of constructing the parking space obstacle judgment area in step S2 includes:

[0021] Select a path point located at a preset distance in front of the vehicle from the local patrol path as the current path reference point, and determine the effective path end reference point;

[0022] The distance between the vehicle and the path is calculated based on the vehicle's current location and the starting point of the local patrol path. When the distance between the vehicle and the path is less than a preset threshold, it is determined that the vehicle is near the current patrol path.

[0023] Starting from the current path reference point, the validity of path points in the local patrol path is statistically analyzed. When the length of the valid path reaches a preset threshold, the parking space obstacle judgment area is generated based on the current path reference point and the end reference point of the valid path.

[0024] In some embodiments of the present invention, the parking space obstacle detection area is a quadrilateral area, and the step of generating the quadrilateral area includes:

[0025] Convert the current path reference point and the effective path end reference point from latitude and longitude coordinates to local map coordinates;

[0026] Calculate the path heading angle based on the current path reference point and the effective path end reference point;

[0027] Based on the path heading angle and the preset lateral translation distance, the system translates along the path direction and the normal vector direction to generate a quadrilateral parking space obstacle judgment area.

[0028] In some embodiments of the present invention, the step of calculating the obstacle point cloud scale within the obstacle judgment area of ​​the parking space in step S3 includes:

[0029] Voxel filtering downsampling is performed on the acquired obstacle point cloud data;

[0030] Transform the filtered point cloud from the vehicle coordinate system to the map coordinate system;

[0031] Traverse the obstacle point cloud in the map coordinate system and determine whether each point is located within the obstacle judgment area of ​​the parking space;

[0032] The number of point clouds located within the obstacle detection area of ​​the parking space is counted to obtain the number of obstacle point clouds within the parking space area;

[0033] When the number of obstacle point clouds is less than a preset threshold and the number of consecutive times the condition is met exceeds a preset counting threshold, the current parking space is determined to meet the cleaning conditions.

[0034] In some embodiments of the present invention, the step of calculating the distance to the obstacle ahead in the vehicle sweeping direction in step S4 includes:

[0035] The space around the vehicle is divided into multiple directional cones according to a preset angular resolution.

[0036] Calculate the horizontal distance and azimuth angle for each point cloud point, and determine the direction bucket to which the point belongs based on the azimuth angle. Each direction bucket retains only the point closest to the vehicle.

[0037] The angle difference is calculated based on the vehicle heading angle and the path heading angle, and the direction bucket index corresponding to the sweeping direction is determined based on the angle difference.

[0038] By selecting the nearest distance between the direction bucket corresponding to the sweeping direction and its adjacent direction bucket, the distance to the obstacle in front of the vehicle in the sweeping direction is obtained.

[0039] In some embodiments of the present invention, the parking space cleaning status in step S5 includes:

[0040] Status 0: Initial state or cleaning exit state, indicating that the cleaning process for a valid parking space has not yet started;

[0041] Status 1: Path valid status, indicating that the length of the patrol path meets the requirements and the parking space obstacle judgment area has been generated;

[0042] Status 2: Cleaning execution status, indicating that the scale of obstacles in the parking space area meets the cleaning conditions, and the vehicle begins to clean the empty parking space;

[0043] Status 3: Insufficient space ahead. This indicates that the distance to obstacles in the direction the vehicle is sweeping is less than the threshold, and sweeping should not continue.

[0044] In some embodiments of the present invention, the step of dynamically updating the parking space cleaning status includes:

[0045] When in state 0, if the local patrol path is valid and the distance between the vehicle and the path is less than the preset threshold, and the parking space obstacle judgment area is successfully generated, then switch to state 1.

[0046] When in state 1, if the number of obstacle point clouds in the parking space area is less than the preset threshold and the number of consecutive times exceeds the preset counting threshold, then switch to state 2; otherwise, maintain state 1.

[0047] When in state 2, if the distance to the obstacle in front of the vehicle in the sweeping direction is less than the preset threshold and the number of consecutive times exceeds the preset counting threshold, then switch to state 3; otherwise, remain in state 2.

[0048] When in state 3, if the parking space route publishing status is not published, then switch to state 0.

[0049] In some embodiments of the present invention, the step of publishing the parking space cleaning path is also included:

[0050] Based on the local sweeping trajectory, the current ring number, and the current location of the vehicle, the cleaning path for the corresponding parking space is dynamically generated.

[0051] The cleaning path is published to the vehicle control module in real time for trajectory tracking;

[0052] In some embodiments of the present invention, the step of dynamically generating the parking space cleaning path includes:

[0053] Obtain the set of local path trajectory points corresponding to the current location of the vehicle, and convert the trajectory points from latitude and longitude coordinates to two-dimensional coordinates in the map coordinate system;

[0054] Calculate the normal vector of the local trajectory based on the coordinates of the trajectory points in the map coordinate system;

[0055] Based on the current number of concentric circles and the normal vector, the local path trajectory points are translated to form the parking space offset path trajectory points;

[0056] A preset entry point is selected on the local trajectory, and after being translated along the normal direction, it is interpolated and connected with the offset path to generate a complete parking space cleaning path.

[0057] The complete parking space cleaning path is dynamically trimmed, retaining only the trajectory points after the vehicle's current position before being published to the vehicle control module.

[0058] This invention proposes an adaptive empty parking space cleaning method based on point cloud obstacle perception. The core of this method lies in generating parking space judgment areas based on local sweeping paths and jointly driving the parking space cleaning state machine using point cloud region statistics and obstacle distances along the cleaning direction, thereby achieving adaptive control of empty parking space cleaning. Compared with existing technologies, this invention has at least the following advantages:

[0059] First, this invention utilizes point cloud obstacle perception to replace the traditional cleaning judgment method that relies solely on fixed paths or visual markers. It constructs a parking space obstacle judgment area using lidar point cloud data and counts the obstacle point cloud scale within the area in real time. This allows for accurate identification of whether there are obstacles such as vehicles, parking locks, cones, or lost items in the parking space. Consequently, it can dynamically decide whether to begin cleaning based on the actual occupancy status of the parking space, fundamentally avoiding ineffective cleaning of occupied parking spaces and significantly improving the effectiveness of cleaning operations and resource utilization.

[0060] Furthermore, this invention is based solely on the geometric distribution statistics (number of points) of the point cloud rather than semantic classification, eliminating the need to distinguish specific obstacle types. It boasts high computational efficiency, is insensitive to changes in ambient lighting, wear and tear on parking markings, and maintains stable and reliable perception performance even under complex lighting conditions such as underground parking lots. Specifically, through voxel filtering downsampling and quadrilateral region cross product judgment, the processing time for a single frame of point cloud can be controlled within 50ms, meeting the real-time vehicle control requirements.

[0061] Secondly, when constructing the quadrilateral parking space obstacle judgment area, this invention comprehensively introduces multiple joint judgment conditions, including: verification that the local patrol path point sequence is not empty, verification that the effective path length reaches a preset threshold, verification that the distance between the vehicle and the local patrol path is less than a preset threshold, and verification of the validity of the vehicle's global positioning status. Only after all the above conditions are met is the cleaning judgment process allowed to proceed. This effectively avoids false triggering caused by the vehicle deviating from the patrol path, insufficient path length, or positioning failure, ensuring that the cleaning decision logic is activated only when the vehicle is in a normal operating posture and there is an effective parking space ahead. Through precise identification of parking space areas using path attribute markers and dynamic statistics of the effective path length, it can accurately distinguish between "path segments with parking spaces" and "path segments without parking spaces," significantly enhancing the scenario-specificity and reliability of the cleaning trigger decision and avoiding false triggering of the cleaning process in non-parking space areas.

[0062] Third, this invention accurately determines the cleaning direction by the difference between the vehicle's heading angle and the path's heading angle, and detects the distance to obstacles ahead in the cleaning direction in real time based on the nearest point calculation method of the direction barrels. Specifically, the 360° space around the vehicle is divided into 360 direction barrels with a resolution of 1°. Each direction barrel retains only the distance to the nearest obstacle. The direction barrel index corresponding to the cleaning direction is determined by combining the heading angle difference, and the shortest distance between adjacent direction barrels is used as the distance judgment value for obstacles ahead. This ensures both detection accuracy and improves the robustness of distance judgment. This invention does not rely on a high-performance computing platform or a large amount of training data, and can respond in real time to dynamic targets ahead of the cleaning path (such as suddenly entering vehicles, pedestrians, temporary piles of objects, etc.). When the distance is less than a preset threshold (such as 1.6m) and a set number of frames (such as 3 frames) are met consecutively, a state switch is triggered in a timely manner to avoid collision accidents and provide effective safety assurance for cleaning operations.

[0063] Fourth, this invention employs a four-state (state 0~state 3) state machine architecture for cleaning state management and introduces a continuous multi-frame counting confirmation mechanism—any state switching condition must be met consecutively a preset number of times before actual switching is executed. This effectively suppresses state jitter and misjudgment caused by single-frame point cloud noise, instantaneous positioning fluctuations, or temporary path anomalies, improving the system's operational stability in complex dynamic environments. The transition logic between each state is clear and complete, forming a closed-loop control link of "initial → valid path → cleaning execution → insufficient space ahead → exit / reset". In actual operation, the state machine can autonomously make decisions based on environmental changes: when the obstacle point cloud size is greater than a threshold, it remains in the valid path state without triggering cleaning; when an obstacle suddenly approaches during cleaning, it autonomously switches to the insufficient space ahead state and suspends cleaning; after the obstacle leaves, it publishes the change in the parking space path state to return the system to the initial state, waiting for the next cleaning trigger condition to be met. This mechanism enables the cleaning vehicle to have human-like environmental adaptive decision-making capabilities, significantly superior to the single fixed cleaning mode in existing technologies.

[0064] Fifth, this invention does not rely on complex deep learning models, large-scale labeled training data, or high-performance GPU computing platforms. The entire technical solution is based on mature point cloud geometric processing (voxel filtering, coordinate transformation, cross product judgment), directional bucket distance statistics, and threshold state machines. It has low computational load and low memory consumption, and can run stably in real time on embedded ARM architecture or low-power x86 platforms. Compared with vision-based deep learning solutions, this invention does not require high graphics card costs and model training overhead, significantly reducing device hardware costs and system deployment complexity.

[0065] Sixth, this invention is applicable to low-speed unmanned floor scrubbers, sweepers, and other sanitation vehicles, and can work stably in various scenarios such as large underground parking lots, shopping mall parking areas, residential parking spaces, and outdoor public parking areas. Through the dynamic offset path automatic generation function (which shifts sequentially based on the current circle number n and the normal vector, with an offset of 0.50m when n=1 and 1.00m when n=2), multi-circle reciprocating cleaning of parking spaces can be achieved, ensuring thorough cleaning without blind spots.

[0066] In summary, this invention effectively addresses the technical pain points of existing unmanned parking space cleaning technologies mentioned in the background section, such as insufficient obstacle perception granularity, rigid cleaning strategies, delayed state switching, and high system costs. It achieves significant improvements in cleaning safety, environmental adaptability, operational efficiency, and system economy, providing a practical and feasible technical solution for the empty parking space cleaning process in the smart sanitation system. Attached Figure Description

[0067] Figure 1 This is a flowchart of the adaptive empty parking space cleaning method based on point cloud obstacle perception of the present invention;

[0068] Figure 2 This is a schematic diagram of the overall control process of the present invention;

[0069] Figure 3 This is a structural diagram of the system of the present invention;

[0070] Figure 4 This is a schematic diagram (A) illustrating the deployment scenario of the present invention's method for cleaning empty parking spaces;

[0071] Figure 5 This is a schematic diagram (B) illustrating the deployment scenario of the present invention's method for cleaning empty parking spaces;

[0072] Figure 6 This is a schematic diagram (C) illustrating the deployment scenario of the present invention's method for cleaning empty parking spaces. Detailed Implementation

[0073] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0074] The following is for reference. Figures 1-6 The adaptive empty parking space cleaning method based on point cloud obstacle perception according to embodiments of the present invention includes the following steps:

[0075] S1. Acquire local patrol path, vehicle pose, positioning status, and point cloud data of obstacles around the vehicle for low-speed unmanned vehicles.

[0076] S2. Based on the local patrol path and vehicle position, construct the obstacle judgment area for parking spaces;

[0077] S3. Calculate the obstacle point cloud size within the obstacle judgment area of ​​the parking space, and determine whether the current parking space meets the cleaning conditions based on the obstacle point cloud size.

[0078] S4. Calculate the distance to obstacles ahead in the sweeping direction of the vehicle, and determine whether there is a risk of continuing to move forward during the sweeping process based on the distance to obstacles ahead;

[0079] S5. Based on path validity, obstacle point cloud size, distance to obstacles ahead, and parking space path release status, dynamically update the parking space cleaning status to achieve adaptive cleaning control of empty parking spaces.

[0080] In some embodiments of the present invention, the step of obtaining the local scanning path in step S1 includes:

[0081] Read the global patrol path, obtain the serial number of the vehicle's current location in the global patrol path, and publish a local patrol path with a preset number of nearby points following the current serial number.

[0082] The local patrol path is a sequence of equally spaced path points.

[0083] Specifically, the process of creating the patrol route can be as follows:

[0084] First, a parking lot environment map is constructed: Real-time data collection of vehicle GPS location, odometer data, and LiDAR environmental perception data is used to identify road boundaries, parking space areas, pillar areas, and obstacle areas within the underground parking lot. A self-developed raster map creation app is used to process the collected data, generating a two-dimensional raster map of the parking lot, providing basic map information for subsequent patrol route planning. Next, lane areas, parking space areas, charging areas, trash can areas, speed bump areas, and no-entry areas are marked on the generated raster map, and corresponding map attribute information is established. Each map area is associated with a unique identifier for subsequent route planning and operational status assessment.

[0085] Second, create a global patrol path on the raster map: Under the raster map (Map), a closed-loop patrol path covering the entire parking lot is formed through manual data collection or automatic generation. Simultaneously, parking space markers are set at the trajectory points. When a vehicle travels to the corresponding trajectory segment, the system can identify whether there is a parking space area on the right side of the current path, thus triggering the parking space detection process. The completed global patrol trajectory points are saved as a global path txt file.

[0086] Third, publish the local patrol path: Read the global path txt file and generate the global patrol path / globalpath. Then, obtain the vehicle's current location's sequence number / id in / globalpath, and publish the local patrol path / nearpath, which consists of 50 neighboring points counting forward from the current sequence number. The local patrol path is preferably a sequence of equally spaced path points, with an adjacent path point spacing of 0.1m, and the total number of path points can be set to 50. Figure 4 , Figure 5 and Figure 6 As shown, the black dashed arrows represent the global patrol path, with the red section indicating the area on the right without parking spaces. The green dotted arrows represent the local patrol path.

[0087] Before cleaning the parking spaces, preparatory work needs to be carried out in the preparatory stage of the cleaning operation, as follows:

[0088] First, adaptive cleaning judgment parameters for path determination, obstacle detection, and state switching are preset, which may include the following parameters:

[0089] The remaining length threshold of the patrol path, PATH_SIZE_THRESHOLD, is used to determine whether there is a sufficiently long effective patrol path ahead of the current vehicle. The preferred value is 24, which corresponds to a path length of approximately 2.4m (the standard width of a small perpendicular parking space).

[0090] The threshold value for the number of parking space obstacle point clouds is OBSTACLE_PCCOUNT_THRESHOLD, which is used to determine whether the size of obstacles in the parking space area meets the cleaning conditions. The preferred value is 25.

[0091] The obstacle distance threshold DISTANCE_FRONT_THRESHOLD is used to determine whether there are obstacles in front of the vehicle in the sweeping direction that may affect the continuation of sweeping. The preferred value is 1.6m.

[0092] The lateral translation distance of the judgment area, DISTANCE_TRANSLATION, is used to generate the parking space obstacle judgment area based on the patrol path. The preferred value is 4.00m.

[0093] The vehicle-path distance threshold DISTANCE_PATH_THRESHOLD is used to determine whether a vehicle is close to the current local patrol path, with a preferred value of 1.20m.

[0094] The preferred size for the point cloud voxel filter leaf is 0.20m × 0.20m × 0.20m.

[0095] The obstacle distance detection angle resolution is preferably set to 1°, corresponding to 360 directional angle barrels;

[0096] The effective range of point cloud distance is preferably set to a minimum distance of 0.30m and a maximum distance of 6.00m;

[0097] A continuous state satisfaction threshold is set to avoid misjudgment in a single frame. It is preferred to set the state to switch after satisfying the threshold three or more times.

[0098] The above parameters can be adjusted according to vehicle size, parking space size, sweeping device width, sensor installation location, and actual operating scenario.

[0099] Secondly, during the parking space cleaning operation, the low-speed autonomous vehicle acquires the following information in real time to reflect the vehicle's operating status, path information, and obstacle point cloud data:

[0100] Obtain the vehicle's current position and heading angle. The vehicle's current position includes longitude and latitude information, and the heading angle is used to indicate the vehicle's current direction of travel.

[0101] Obtain the reference origin required for map coordinate transformation. The reference origin includes reference longitude and reference latitude.

[0102] Obtain the vehicle's global positioning status. When the positioning status meets the preset positioning validity conditions, allow the subsequent obstacle perception and cleaning status judgment process.

[0103] Obtain the local patrol path, which consists of multiple path points. Each path point contains location coordinates and path attribute flags.

[0104] Acquire point cloud data of obstacles around the vehicle. The obstacle point cloud data is used to determine the size of obstacles in the parking space area and the distance of obstacles in front of the vehicle.

[0105] Get the parking space path publishing status. The parking space path publishing status is used to determine whether the current parking space cleaning task is still in a valid state.

[0106] Furthermore, the effectiveness of the local patrol path is checked. The specific checking process is as follows:

[0107] 1. After receiving the local patrol path, first determine whether the path point sequence is empty. If the path point sequence is empty, do not perform parking space cleaning judgment, and output a prompt message that the path is empty.

[0108] 2. When the path point sequence is not empty, select a path point located at a preset distance in front of the vehicle in the local sweeping path as the current path reference point. Preferably, the 10th path point is selected as the current path reference point to achieve sweeping judgment approximately 1m in advance.

[0109] 3. Calculate the distance between the vehicle and the path based on the vehicle's current location and the starting point of the local patrol path. Specifically, convert the vehicle's current location and the starting point of the path from latitude and longitude coordinates to local planar map coordinates, and calculate the Euclidean distance between the two points to obtain the path distance (path_distance). When path_distance is less than the vehicle-path distance threshold, it is determined that the vehicle is near the current patrol path.

[0110] 4. Starting from the current path reference point, perform validity statistics on path points in the local patrol path. When the path attribute flag of a path point meets the preset valid path flag, it is included in the valid path length; when the path attribute flag does not meet the preset valid path flag, the statistics stop. Preferably, when the orientation.y of a path point equals 2 after being converted to an integer, the path point is determined to be a valid patrol path point.

[0111] 5. Set the initial path count to 10, and start counting from the 10th path point up to the remaining length threshold of the sweeping path. When the number of valid path points counted is less than a preset proportion, it is determined that the current sweeping path length is insufficient, and the parking space cleaning status is set to the initial state. Preferably, when the number of valid path points is less than PATH_SIZE_THRESHOLD / 2 (half the width of a parking space), it is determined that the sweeping path length is insufficient.

[0112] like Figure 4 and Figure 5 As shown in the diagram, when the floor cleaning vehicle moves to positions A and B, the effective path length is greater than the remaining sweeping path length threshold, and the distance between the vehicle and the local sweeping path is less than the vehicle-path distance threshold. At this time, / parking_status (parking space cleaning status) switches to 1, indicating that the sweeping path meets the conditions. Figure 6 As shown, when the floor cleaning vehicle moves to position C, there is no parking space marker on this path, and the effective path length is 0, which does not meet the threshold for the remaining length of the sweeping path. Therefore, / parking_status is switched to 0 here.

[0113] In some embodiments of the present invention, the step of constructing the parking space obstacle judgment area in step S2 includes:

[0114] Select a path point located at a preset distance in front of the vehicle from the local patrol path as the current path reference point, and determine the effective path end reference point;

[0115] The distance between the vehicle and the path is calculated based on the vehicle's current location and the starting point of the local patrol path. When the distance between the vehicle and the path is less than a preset threshold, it is determined that the vehicle is near the current patrol path.

[0116] Starting from the current path reference point, the validity of path points in the local patrol path is statistically analyzed. When the length of the valid path reaches a preset threshold, a parking space obstacle judgment area is generated based on the current path reference point and the end reference point of the valid path.

[0117] For example, a parking space obstacle judgment area is generated when the local patrol path meets the following conditions: 1. The effective path length reaches the remaining length threshold of the patrol path; 2. The distance between the vehicle and the local patrol path is less than the vehicle-path distance threshold; 3. The current parking space cleaning status is the initial state or the effective path state.

[0118] Specifically, the obstacle detection area for parking spaces can be a quadrilateral area, and its generation steps may include:

[0119] Convert the current path reference point and the effective path end reference point from latitude and longitude coordinates to local map coordinates;

[0120] Calculate the path heading angle based on the current path reference point and the effective path end reference point;

[0121] Based on the path heading angle and the preset lateral translation distance, the system translates along the path direction and the normal vector direction to generate a quadrilateral parking space obstacle judgment area.

[0122] Specifically, the process of generating the obstacle detection area for parking spaces may include:

[0123] 1) Convert the current path reference point and the effective path end reference point from latitude and longitude coordinates to local map coordinates. Assume the coordinates corresponding to the current path reference point are... The coordinates of the reference point at the end of the effective path are: .

[0124] The conversion relationship from latitude and longitude to local map coordinates is as follows:

[0125] ,

[0126] in, and These are the latitude and longitude of the point to be converted, respectively. and These are the latitude and longitude of the reference origin, respectively. The radius of the Earth is 6378137.0m.

[0127] 2) Calculate the path heading angle based on the current path reference point and the effective path end reference point. The path heading angle is defined as 0° for true north and positive for clockwise direction. The calculation formula is:

[0128] ,

[0129] in, This is the difference between the second path point and the first path point in the longitude direction or the local lateral direction. This is the difference between the second path point and the first path point in the latitudinal or local longitudinal direction; when the calculated result is less than 0°, it is added by 360° to keep the path heading angle range within [0°, 360°).

[0130] 3) Generate the parking space obstacle judgment area based on the path direction and the lateral translation distance of the judgment area, specifically including:

[0131] First, according to and Calculate the path direction vector and normalize it.

[0132] Secondly, the starting point is shifted a preset distance along the path to obtain the forward boundary point of the judgment area;

[0133] Next, calculate the normal vector of the path direction, and translate along the direction of the normal vector. Generate the other two vertices of the parking space determination area;

[0134] Finally, the above points form a quadrilateral parking space obstacle judgment area.

[0135] Understandably, the aforementioned quadrilateral judgment area is used to represent the parking space area that the vehicle is about to clean. After generating the judgment area, the parking space cleaning status is updated to a valid path status, and the quadrilateral judgment area can be published as a visual marker for debugging and operation monitoring.

[0136] like Figure 4 and Figure 5 As shown in the diagram, when the floor scrubbing vehicle moves to positions A and B, the current parking space cleaning status is a valid path state. At this time, a parking space obstacle detection area will be generated, such as... Figure 4 and Figure 5 The area within the red dotted-line box. For example... Figure 6 As shown, when the floor cleaning vehicle moves to position C, there are no parking space markers on this path, and the effective path length is 0, which does not meet the threshold for the remaining length of the sweeping path. Therefore, no parking space obstacle judgment area will be generated here.

[0137] In some embodiments of the present invention, the step of calculating the obstacle point cloud size within the parking space obstacle judgment area in step S3 may include:

[0138] Voxel filtering downsampling is performed on the acquired obstacle point cloud data;

[0139] Transform the filtered point cloud from the vehicle coordinate system to the map coordinate system;

[0140] Traverse the obstacle point cloud in the map coordinate system and determine whether each point is located within the parking space obstacle detection area;

[0141] The number of point clouds located within the obstacle detection area of ​​the parking space is counted to obtain the number of obstacle point clouds within the parking space area;

[0142] When the number of obstacle point clouds is less than a preset threshold and the number of consecutive times the condition is met exceeds a preset counting threshold, the current parking space is determined to meet the cleaning conditions.

[0143] Specifically, the statistical analysis of the obstacle point cloud size within the obstacle detection area of ​​a parking space may include the following steps:

[0144] S31. After receiving the point cloud data of obstacles around the vehicle, the point cloud data can be converted from message format to point cloud processing format and voxel filtering downsampling can be performed; wherein, the leaf size of the voxel filter is preferably 0.20 × 0.20 × 0.20 m, so as to reduce the amount of point cloud data and improve the efficiency of subsequent obstacle perception.

[0145] S32. Since the obstacle detection area is located in the map coordinate system, while the original point cloud data is located in the vehicle coordinate system, it is necessary to obtain the transformation relationship from the vehicle coordinate system to the map coordinate system. Based on the translation vector and rotation quaternion from the vehicle coordinate system to the map coordinate system, a rigid body transformation matrix can be constructed, and the filtered point cloud can be transformed from the vehicle coordinate system to the map coordinate system to obtain the obstacle point cloud in the map coordinate system.

[0146] S33. When the vehicle positioning status meets the valid conditions and the parking space cleaning status is not the initial state, perform subsequent obstacle size calculation and obstacle distance calculation in front of the vehicle within the parking space area.

[0147] S34. Traverse the obstacle point cloud in the map coordinate system, project each point as a planar point, and determine whether the point is located within the parking space obstacle judgment area of ​​the quadrilateral. For example, when determining whether a point is inside the quadrilateral, the cross product method can be used. For each of the four sides of the quadrilateral, calculate the cross product between the point to be judged and the edge vector. When the vertices of the quadrilateral are arranged in a preset order, if the point to be judged is located on the outside of any side, it is determined that the point is not inside the quadrilateral. If the point to be judged is located on the inside of all sides, it is determined that the point is located within the parking space obstacle judgment area.

[0148] S35. Count the number of point clouds located within the quadrilateral judgment area to obtain the number of obstacle point clouds within the parking space area. This parameter is used to characterize the size of obstacles within the current parking space area.

[0149] S36, when Less than the threshold for the number of obstacle point clouds in parking spaces If the number of consecutive occurrences exceeds the preset counting threshold, it is determined that the obstacle size in the current parking space area is small, and the conditions for starting cleaning are met.

[0150] like Figure 4 As shown, when the floor scrubbing vehicle is at position A, the number of point clouds in the obstacle detection area of ​​the parking space on the right is 0, which is less than the threshold for the number of point clouds of parking space obstacles. When the condition is met for more than 3 consecutive times, the current parking space area is determined to meet the conditions to start cleaning, and the parking space cleaning status is changed. Switching to 2 indicates that you are ready to begin cleaning the parking space. For example... Figure 5 As shown, when the floor scrubbing vehicle is at position B, the number of point clouds in the obstacle detection area of ​​the parking space on the right exceeds the threshold for the number of point clouds in the parking space obstacle. Therefore, it is determined that the current parking space area does not meet the conditions for starting cleaning, and the parking space cleaning status is changed. Switching to 0 indicates that cleaning has not started.

[0151] In some embodiments of the present invention, the step of publishing the parking space cleaning path is also included:

[0152] Based on the local sweeping trajectory, the current ring number, and the current location of the vehicle, the cleaning path for the corresponding parking space is dynamically generated.

[0153] The cleaning path is published to the vehicle control module in real time for trajectory tracking.

[0154] Specifically, the dynamic generation of parking space cleaning paths may include the following steps:

[0155] T1. Parking space offset trajectory construction:

[0156] First, obtain the set of local path trajectory points corresponding to the vehicle's current location, and transform the trajectory points from latitude and longitude coordinates to two-dimensional coordinates in the map coordinate system. Let the WGS84 latitude and longitude coordinates of the i-th trajectory point be represented as... Simultaneously obtain the current vehicle GPS location. The WGS84 latitude and longitude coordinates corresponding to the origin of the raster map. And the current cleaning cycle number n. (Based on the origin of the raster map Map) Using local coordinate transformation reference points, a coordinate transformation from WGS84 to the local Map coordinate system is performed on the local path trajectory points and the current vehicle GPS position to obtain the two-dimensional coordinates of each trajectory point in the Map coordinate system. And the current two-dimensional coordinates of the vehicle in the Map coordinate system. .

[0157] The formula for converting trajectory points is:

[0158] ,

[0159] Wherein, R = 6378137.0m.

[0160] The above conversion enables a unified representation between geographic coordinates and local map coordinates, providing a foundation for subsequent trajectory calculations.

[0161] Secondly, based on the coordinates of the trajectory points in the map coordinate system (i.e., the map coordinate system) Calculate the normal vector of the local trajectory to prepare for subsequent translation.

[0162] adjacent point direction The calculation formula is: ;

[0163] The trajectory length is Unit tangent vector: ;

[0164] Calculate the normal vector using the right-hand rule: ;

[0165] The normal vector represents the lateral offset direction of the trajectory point perpendicular to the driving direction, and is used for subsequent parking space trajectory generation.

[0166] T2. Generate parking space offset path and connection path:

[0167] First, based on the current circle number n and the normal vector, the local path trajectory points are translated to form the parking space offset path trajectory points. :

[0168]

[0169] in, This refers to the translation distance. Since the width of the currently used unmanned floor scrubber is 1.3m, to ensure effective cleaning, it is necessary to... This method can automatically generate corresponding parking space offset paths based on different cleaning cycles.

[0170] Then select a preset entry point on the local trajectory. Translate along the normal direction to obtain Then, linear interpolation is performed based on the distance between the two points to obtain the connecting trajectory points, so that the vehicle can smoothly transition from the local patrol path to the parking space offset path.

[0171] T3. Publish the complete parking space cleaning trajectory:

[0172] The connecting and offset trajectories are stitched together to form a complete parking space cleaning path. Simultaneously, each trajectory point in the path is assigned a unique trajectory ID according to its generation order, used for subsequent nearest-point search and path trimming. The current vehicle GPS position is converted to local coordinates in the Map coordinate system, and the distance between the vehicle and each trajectory point in the complete path is calculated. The trajectory point with the smallest distance is found. To prevent GPS jitter from causing trajectory regression, a historical nearest-point ID constraint is introduced, allowing only monotonically increasing trajectory IDs, and subsequently only the trajectory after the nearest point is retained. After the complete parking space cleaning path is converted back from Map coordinates to latitude and longitude coordinates, it is published to the vehicle control module in real time. This dynamic trimming mechanism avoids the vehicle repeatedly tracking already traversed paths, improving the stability and efficiency of path tracking during parking space cleaning.

[0173] For example, when the floor cleaning truck is at location A and determines that the current parking space area meets the conditions for starting cleaning, it publishes the parking space cleaning trajectory. ,like Figure 4 As shown by the blue dotted arrow in the image.

[0174] In some embodiments of the present invention, the step of calculating the distance to the obstacle ahead in the vehicle sweeping direction in step S4 includes:

[0175] The space around the vehicle is divided into multiple directional cones according to a preset angular resolution.

[0176] Calculate the horizontal distance and azimuth angle for each point cloud point, determine the direction bucket to which the point belongs based on the azimuth angle, and retain only the point closest to the vehicle in each direction bucket;

[0177] Calculate the angle difference between the vehicle's heading angle and the path's heading angle, and determine the direction bucket index corresponding to the sweeping direction based on the angle difference;

[0178] By selecting the nearest distance between the direction bucket corresponding to the sweeping direction and its adjacent direction bucket, the distance to the obstacle in front of the vehicle in the sweeping direction is obtained.

[0179] Specifically, to determine whether there are obstacles in front of the vehicle that could affect continued cleaning during the parking space cleaning process, the filtered point cloud in the vehicle coordinate system is divided into directions and the nearest point is extracted. The calculation of the distance to obstacles in the vehicle's cleaning direction may include the following steps:

[0180] S41. First, divide the 360° space around the vehicle into multiple steering barrels according to a preset angle resolution; wherein, the preferred angle resolution is 1° and the number of steering barrels is 360.

[0181] S42. Calculate the horizontal distance for each point cloud point. and azimuth , of which, horizontal distance It filters out points with a distance less than 0.30m or greater than 6.00m; simultaneously, the azimuth angle... And convert the angle to [0, 2π); determine the direction bucket to which the point belongs based on the azimuth angle, and retain only the point closest to the vehicle for each direction bucket to obtain the distance to the nearest obstacle in each direction;

[0182] S43, based on the vehicle heading angle With path heading angle Calculate the angle difference : And based on the angle difference Determine the direction bucket index corresponding to the cleaning direction. : ;

[0183] S44, Select The distance to the nearest obstacle in the vehicle's sweeping direction is obtained by measuring the distance in the adjacent buckets and the nearest distance in the adjacent buckets. Preferably, the following can be selected: , , , , The minimum closest distance among the 5 directional buckets is used as To improve the robustness of obstacle distance judgment.

[0184] In some embodiments of the present invention, the parking space cleaning status in step S5 includes:

[0185] Status 0: Initial state or cleaning exit state, indicating that the cleaning process for a valid parking space has not yet started;

[0186] Status 1: Path valid status, indicating that the length of the patrol path meets the requirements and the parking space obstacle judgment area has been generated;

[0187] Status 2: Cleaning execution status, indicating that the scale of obstacles in the parking space area meets the cleaning conditions, and the vehicle begins to clean the empty parking space;

[0188] Status 3: Insufficient space ahead. This indicates that the distance to obstacles in the direction the vehicle is sweeping is less than the threshold, and sweeping should not continue.

[0189] Specifically, the rules for transitioning between states, i.e., the steps for dynamically updating the parking space cleaning status, may include:

[0190] When in state 0, if the local patrol path is valid and the distance between the vehicle and the path is less than the preset threshold, and the parking space obstacle judgment area is successfully generated, then switch to state 1.

[0191] When in state 1, if the number of obstacle point clouds in the parking space area is less than the preset threshold and the number of consecutive times exceeds the preset counting threshold, then switch to state 2, indicating that the parking space cleaning will begin; otherwise, remain in state 1.

[0192] When in state 2, if the distance to the obstacle in front of the vehicle in the sweeping direction is less than the preset threshold and the number of consecutive times exceeds the preset counting threshold, then switch to state 3, indicating that there is not enough space in front and it is not advisable to continue sweeping; otherwise, maintain state 2.

[0193] When in state 3, if the parking space path publishing status is not published, then switch to state 0, indicating that the current parking is the end of the cleaning process or exit.

[0194] Understandably, the aforementioned state mechanism can prevent misjudgments caused by single-frame point cloud noise, instantaneous positioning fluctuations, or temporary path anomalies, thereby improving the stability and reliability of parking space cleaning state switching.

[0195] Therefore, after completing path determination, obstacle point cloud perception, and state machine update, the parking space cleaning status is released. The parking space cleaning status is used to control whether the low-speed unmanned floor cleaning vehicle enters the parking space for cleaning, whether to continue cleaning, and whether to stop cleaning.

[0196] When the parking space cleaning status is 0, the vehicle remains in a non-cleaning or exited state; when the parking space cleaning status is 1, the vehicle enters the parking space cleaning preparation state; when the parking space cleaning status is 2, the vehicle performs parking space cleaning; when the parking space cleaning status is 3, the vehicle stops cleaning or waits for the upper-level control module to re-plan.

[0197] In summary, the adaptive empty parking space cleaning method based on point cloud obstacle perception provided by this invention is based on the construction of an integrated intelligent cleaning control framework encompassing "perception-judgment-decision-execution". This method first fuses multi-source sensor data such as GPS, LiDAR, and odometer to create a parking lot grid map and generate a global patrol path. Then, it dynamically publishes equally spaced local patrol paths based on the real-time vehicle location, providing a precise path basis for triggering cleaning. After confirming the path's validity, the method automatically generates a quadrilateral parking space obstacle judgment area along the patrol path direction. By transforming the LiDAR point cloud from the vehicle coordinate system to the map coordinate system and using the cross product method to perform regional point cloud statistics, it accurately perceives the spatial distribution and scale of obstacles within the parking space, thereby identifying available parking spaces. During the cleaning execution phase, the method divides the 360-degree space around the vehicle into multiple directional cones, uses the angular difference between the vehicle's heading and the path's heading to determine the cleaning direction, and calculates the closest distance to obstacles directly in front in real time, enabling the vehicle to quickly respond to sudden intrusions (such as pedestrians or other vehicles). Building upon this foundation, a four-state machine is constructed, encompassing initial state, path validity state, cleaning execution state, and insufficient space ahead state. Path validity, point cloud statistical scale, distance to obstacles ahead, and path deployment status serve as joint driving signals, combined with a continuous multi-frame counting confirmation mechanism to achieve smooth and reliable switching of cleaning states. The entire method does not rely on deep learning models or high-performance computing platforms; it completes the entire process decision-making solely through point cloud geometry processing, coordinate transformation, and a threshold state machine, offering advantages such as lightweight computation, low cost, and strong robustness to illumination. Experimental and application results demonstrate that this method can achieve safe and efficient adaptive cleaning of empty parking spaces in various complex scenarios such as underground parking lots, commercial complexes, and residential communities. It effectively solves the technical challenges of coarse perception granularity, rigid strategies, lag response, and high deployment costs inherent in traditional solutions, providing practical technical support for smart sanitation systems.

[0198] Although embodiments of the invention have been shown and described, those skilled in the art will understand 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 claims and their equivalents.

Claims

1. An adaptive empty parking space cleaning method based on point cloud obstacle perception, characterized in that, Includes the following steps: S1. Acquire local patrol path, vehicle pose, positioning status, and point cloud data of obstacles around the vehicle for low-speed unmanned vehicles. S2. Based on the local patrol path and the vehicle position, construct the parking space obstacle judgment area; S3. Calculate the obstacle point cloud size within the obstacle judgment area of ​​the parking space, and determine whether the current parking space meets the cleaning conditions based on the obstacle point cloud size. S4. Calculate the distance to obstacles ahead in the sweeping direction of the vehicle, and determine whether there is a risk of continuing to move forward during the sweeping process based on the distance to obstacles ahead; S5. Based on path validity, obstacle point cloud size, distance to obstacles ahead, and parking space path publishing status, dynamically update parking space cleaning status to achieve adaptive cleaning control of empty parking spaces.

2. The adaptive empty parking space cleaning method based on point cloud obstacle perception according to claim 1, characterized in that, The step of obtaining the local patrol path in step S1 includes: Read the global patrol path, obtain the serial number of the vehicle's current location in the global patrol path, and publish the local patrol path with a preset number of nearby points following the current serial number; The local patrol path is a sequence of equally spaced path points.

3. The adaptive empty parking space cleaning method based on point cloud obstacle perception according to claim 1, characterized in that, The step of constructing the obstacle detection area for the parking space in step S2 includes: Select a path point located at a preset distance in front of the vehicle from the local patrol path as the current path reference point, and determine the effective path end reference point; The distance between the vehicle and the path is calculated based on the vehicle's current location and the starting point of the local patrol path. When the distance between the vehicle and the path is less than a preset threshold, it is determined that the vehicle is near the current patrol path. Starting from the current path reference point, the validity of path points in the local patrol path is statistically analyzed. When the length of the valid path reaches a preset threshold, the parking space obstacle judgment area is generated based on the current path reference point and the end reference point of the valid path.

4. The adaptive empty parking space cleaning method based on point cloud obstacle perception according to claim 3, characterized in that, The obstacle detection area for the parking space is a quadrilateral area, and the steps for generating the quadrilateral area include: Convert the current path reference point and the effective path end reference point from latitude and longitude coordinates to local map coordinates; Calculate the path heading angle based on the current path reference point and the effective path end reference point; Based on the path heading angle and the preset lateral translation distance, the system translates along the path direction and the normal vector direction to generate a quadrilateral parking space obstacle judgment area.

5. The adaptive empty parking space cleaning method based on point cloud obstacle perception according to claim 1, characterized in that, The step S3, which involves calculating the obstacle point cloud size within the obstacle judgment area of ​​the parking space, includes: Voxel filtering downsampling is performed on the acquired obstacle point cloud data; Transform the filtered point cloud from the vehicle coordinate system to the map coordinate system; Traverse the obstacle point cloud in the map coordinate system and determine whether each point is located within the obstacle judgment area of ​​the parking space; The number of point clouds located within the obstacle detection area of ​​the parking space is counted to obtain the number of obstacle point clouds within the parking space area; When the number of obstacle point clouds is less than a preset threshold and the number of consecutive times the condition is met exceeds a preset counting threshold, the current parking space is determined to meet the cleaning conditions.

6. The adaptive empty parking space cleaning method based on point cloud obstacle perception according to claim 1, characterized in that, The step S4, which calculates the distance to obstacles ahead in the vehicle's sweeping direction, includes: The space around the vehicle is divided into multiple directional cones according to a preset angular resolution. Calculate the horizontal distance and azimuth angle for each point cloud point, and determine the direction bucket to which the point belongs based on the azimuth angle. Each direction bucket retains only the point closest to the vehicle. The angle difference is calculated based on the vehicle heading angle and the path heading angle, and the direction bucket index corresponding to the sweeping direction is determined based on the angle difference. By selecting the nearest distance between the direction bucket corresponding to the sweeping direction and its adjacent direction bucket, the distance to the obstacle in front of the vehicle in the sweeping direction is obtained.

7. The adaptive empty parking space cleaning method based on point cloud obstacle perception according to claim 1, characterized in that, The parking space cleaning status in step S5 includes: Status 0: Initial state or cleaning exit state, indicating that the cleaning process for a valid parking space has not yet started; Status 1: Path valid status, indicating that the length of the patrol path meets the requirements and the parking space obstacle judgment area has been generated; Status 2: Cleaning execution status, indicating that the scale of obstacles in the parking space area meets the cleaning conditions, and the vehicle begins to clean the empty parking space; Status 3: Insufficient space ahead, indicating that the distance to the obstacle in the direction the vehicle is sweeping is less than the threshold, and sweeping should not continue.

8. The adaptive empty parking space cleaning method based on point cloud obstacle perception according to claim 7, characterized in that, The steps for dynamically updating the cleaning status of the parking space include: When in state 0, if the local patrol path is valid and the distance between the vehicle and the path is less than the preset threshold, and the parking space obstacle judgment area is successfully generated, then switch to state 1. When in state 1, if the number of obstacle point clouds in the parking space area is less than the preset threshold and the number of consecutive times exceeds the preset counting threshold, then switch to state 2; otherwise, maintain state 1. When in state 2, if the distance to the obstacle in front of the vehicle in the sweeping direction is less than the preset threshold and the number of consecutive times exceeds the preset counting threshold, then switch to state 3; otherwise, remain in state 2. When in state 3, if the parking space route publishing status is not published, then switch to state 0.

9. The adaptive empty parking space cleaning method based on point cloud obstacle perception according to claim 1, characterized in that, It also includes the step of publishing parking space cleaning routes: Based on the local sweeping trajectory, the current ring number, and the current location of the vehicle, the cleaning path for the corresponding parking space is dynamically generated. The cleaning path is published to the vehicle control module in real time for trajectory tracking.

10. The adaptive empty parking space cleaning method based on point cloud obstacle perception according to claim 9, characterized in that, The steps for dynamically generating the cleaning path for the parking space include: Obtain the set of local path trajectory points corresponding to the current location of the vehicle, and convert the trajectory points from latitude and longitude coordinates to two-dimensional coordinates in the map coordinate system; Calculate the normal vector of the local trajectory based on the coordinates of the trajectory points in the map coordinate system; Based on the current number of concentric circles and the normal vector, the local path trajectory points are translated to form the parking space offset path trajectory points; A preset entry point is selected on the local trajectory, and after being translated along the normal direction, it is interpolated and connected with the offset path to generate a complete parking space cleaning path. The complete parking space cleaning path is dynamically trimmed, retaining only the trajectory points after the vehicle's current position before being published to the vehicle control module.

Images