Multi-unmanned aerial vehicle building detection method based on heuristic sampling and clustering division

Abstract

The invention discloses a multi-unmanned aerial vehicle building detection method based on heuristic sampling and clustering division, and the method comprises the following steps: S1, combining a grid map with an octree map, and carrying out the modeling of an environment through considering the search and calculation efficiency; s2, preprocessing the bounding box for accelerating the starting of the unmanned aerial vehicle cluster and limiting the working range of the unmanned aerial vehicle cluster; s3, the exploration unmanned aerial vehicle carries a laser radar to carry out modeling on a scene, all road points are accessed in sequence through a heuristic function, and rough detection is carried out on the surface of an object in the movement process; s4, the multiple detection unmanned aerial vehicles detect targets in respective task areas based on clustering division; s5, the grid map is combined with the octree map, search and calculation efficiency are maximized, and modeling and characterization are conducted on the environment; and S6, detecting map information exchanged among the unmanned aerial vehicles, undetected interest point information and a task division result, and carrying out environment detection along a planned safe path.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer data processing, and in particular relates to a multi-UAV building detection method based on heuristic sampling and clustering. Background Art

[0002] As society develops, more high-rise buildings face disrepair. Large equipment like aircraft also require regular facade inspections. Detecting defects in target facades and providing early warning of potential dangers has become a crucial issue. Given the scale of buildings, conventional inspection methods are time-consuming and labor-intensive. Multi-rotor drones, with their exceptional three-dimensional maneuverability, are ideal for these inspections. Manually controlling drones not only requires professional pilot training, but also makes it difficult to detect missed areas in complex environments, is inefficient, and is costly. Therefore, autonomous drone exploration and inspection represent a significant market opportunity.

[0003] The development of drone technology and applications has led to the emergence of a range of methods for space exploration, target modeling, and detection. Among these, there has been considerable research into exploration solutions using drones equipped with lidar. These tasks involve one or more high-performance drones autonomously exploring an environment. Within a designated mission area, the drones autonomously explore the environment, modeling the target and detecting its surface.

[0004] However, for large buildings or equipment, the insufficient exploration and inspection capabilities of a single drone can lead to wasted time, while deploying multiple high-performance drones simultaneously increases deployment costs. Existing heterogeneous drone swarming solutions rely heavily on pre-calculation, requiring the pre-processing of bounding boxes encompassing the entire target to generate one or more paths, which the drones then simply track. This approach cannot be processed online; it also lacks robustness, making it difficult to handle complex scenarios and relies heavily on bounding boxes; or it is computationally expensive, making it difficult to solve the problem online or onboard. Summary of the Invention

[0005] To solve the above problems, the present invention proposes a multi-UAV building detection method based on heuristic sampling and clustering.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A multi-UAV building detection method based on heuristic sampling and clustering includes the following steps: S1. Use a grid map combined with an octree map to model the environment, taking into account search and computational efficiency. Based on the size and complexity of the map, the map resolution is independently selected to accelerate computation and query. S2, pre-processing the bounding box to accelerate the startup of the drone cluster and limit the working range of the drone cluster; S3: The exploration drone uses a lidar to model the scene. It visits each waypoint sequentially through a heuristic function to fully model the objects of interest in the environment. It also performs a rough inspection of the object surface during motion. The rough inspection results are provided to the detection drone as initial task division data. S4, multiple detection drones detect targets in their respective mission areas based on clustering division; S5. Use grid maps combined with octree maps to maximize search and computational efficiency to model and represent the environment for UAV motion path planning. S6. The detection drones exchange map information, information about undetected points of interest, and task division results, and perform environmental detection along the planned safe path.

[0007] Preferably, in step S1, the environment includes a target building. Initially, all points in the map are set to be occupied, and a safe space is calculated through continuous scanning by the laser radar through line of sight. When the laser radar illuminates a surface within the bounding box, the surface point set is supplemented and all surface points are set to an undetected state.

[0008] Preferably, the specific process of preprocessing the bounding box in step S2 is: S21. Sampling multiple points inside and near the bounding box according to a predetermined rule and marking them with features to accelerate subsequent calculations; wherein the feature markings include whether they are surface points, whether they are interior points, and whether they have been detected; S22. Sample multiple waypoints in the bounding box as candidate waypoints for exploring the drone.

[0009] Preferably, in step S2, the bounding box is a three-dimensional frame containing the object of interest, and the target within the three-dimensional frame is modeled and detected by the drone cluster.

[0010] Preferably, in step S3, for the Waypoints , the heuristic function is: ,in, For the The heuristic value of each waypoint; For the The degree of unexplored and undetected areas near each waypoint; For the The distance between each waypoint and other drones of the same type; To go to The distance between waypoints; The distance to the aircraft that has not met the takeoff conditions is adopted; a reasonable distance strategy is adopted to maintain the obstacle. Based on the current map, the initial waypoint is offset. The offset direction is along the line connecting the current position to the center of gravity of the nearby obstacle, and the position is shifted forward or backward to the optimal observation distance; through communication, multiple exploration drones synchronize the explored waypoints with each other to avoid repeated exploration, and the maps are updated with each other during the communication process to provide an initial rough inspection result; wherein, the rough inspection result is a set of points of interest.

[0011] Preferably, the specific processing process of step S4 is: S41. Based on the location of each detection drone, the initial rough inspection results are clustered based on distance, and each detection drone is assigned a portion of the task area. The task assignment process is restarted after receiving new information greater than a threshold or after a period of time. S42. For each mission area, each detection UAV performs density-based clustering on known points of interest or surface points to be detected within the mission area to refine its own mission area. S43. Visit each task area in turn through a heuristic function to detect the target in the set of interest points.

[0012] Preferably, in step S5, after the exploration drone or detection drone selects the next waypoint, an initial path will be planned in the grid map using the A* path planning algorithm, and the initial path will be dynamically optimized using the MinimumSnap motion planning algorithm; the dynamically feasible path will continue to be tested for safety, and if it is unsafe, it will continue to be optimized.

[0013] Preferably, the exploration drone and the detection drone are both equipped with a gimbal camera to detect the surface of the object, and the posture of the detection drone and the gimbal camera are comprehensively controlled to balance the number of targets in the field of view and the detection quality.

[0014] Preferably, the communication and integration module of the ground station coordinates the communication between the exploration drone and the detection drone, and timestamps the information, and the information that does not participate in the logical calculation is based on the latest timestamp.

[0015] By adopting the above-mentioned technical solution, the present invention has the following beneficial effects: Using heterogeneous drone swarms as a practical and feasible solution, the present invention comprehensively considers deployment costs, detection effectiveness, and detection efficiency, and proposes a distributed, online, airborne computation-enabled method for heterogeneous multi-drone collaborative detection. By assigning different tasks to drones equipped with different equipment and coordinating the mission areas of multiple drones within each type, the present invention enables exploration and detection in unknown spaces using heterogeneous drone swarms, while addressing various practical challenges, such as communication limitations, observation angle limitations, and flight range limitations. The present invention uses some radar-equipped aircraft to model buildings, allowing the remaining drones to utilize this information to conduct exploration and detection without the need for expensive and heavy radar equipment. Furthermore, to address the high complexity of task scheduling associated with heterogeneous drone swarms, the present invention focuses on addressing how to better allocate tasks among drones and how to better execute their respective tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of the present invention; Figure 2 It is a framework diagram of the present invention; Figure 3 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] like Figures 1 to 3 As shown in FIG, a multi-UAV building detection method based on heuristic sampling and clustering includes the following steps: S1. Use a grid map combined with an octree map to model the environment, taking into account search and computational efficiency. Based on the size and complexity of the map, the map resolution is independently selected to accelerate computation and query. In step S1, the environment includes the target building. Initially, all points in the map are set to be occupied. The LiDAR continuously scans and calculates the safe space through line of sight. When the LiDAR sees a surface within the bounding box, the surface point set is replenished and all surface points are set to an undetected state. S2, pre-processing the bounding box to accelerate the startup of the drone cluster and limit the working range of the drone cluster; In step S2, the bounding box is a three-dimensional frame containing the object of interest, and the target within the three-dimensional frame is modeled and detected by the drone cluster; The specific process of preprocessing the bounding box in step S2 is as follows: S21. Sampling multiple points inside and near the bounding box according to a predetermined rule and marking them with features to accelerate subsequent calculations; wherein the feature markings include whether they are surface points, whether they are interior points, and whether they have been detected; S22, sampling multiple waypoints in the bounding box as candidate waypoints for exploring the drone; S3: The exploration drone uses a lidar to model the scene. It visits each waypoint sequentially through a heuristic function to fully model the objects of interest in the environment. It also performs a rough inspection of the object surface during motion. The rough inspection results are provided to the detection drone as initial task division data. In step S3, for Waypoints , the heuristic function is: ,in, For the The heuristic value of each waypoint; For the The degree of unexplored and undetected areas near each waypoint; For the The distance between each waypoint and other drones of the same type; To go to The distance between waypoints; The distance to an aircraft that has not met takeoff conditions is determined by maintaining a reasonable distance from obstacles. Based on the current map, the initial waypoint is offset forward or backward along the line connecting the current position to the center of gravity of nearby obstacles to the optimal observation distance. Through communication, multiple exploration drones synchronize their explored waypoints to avoid repeated exploration, and maps are updated during communication to provide initial rough inspection results, which are a set of points of interest. S4, multiple detection drones detect targets in their respective mission areas based on clustering division; The specific processing process of step S4 is: S41. Based on the location of each detection drone, the initial rough inspection results are clustered based on distance, and each detection drone is assigned a portion of the task area. The task assignment process is restarted after receiving new information greater than a threshold or after a period of time. S42. For each mission area, each detection UAV performs density-based clustering on known points of interest or surface points to be detected within the mission area to refine its own mission area. S43, through the heuristic function, sequentially visiting each task area to detect the target in the set of interest points; S5. Use grid maps combined with octree maps to maximize search and computational efficiency to model and represent the environment for UAV motion path planning. In step S5, after the exploration drone or detection drone selects the next waypoint, it will use the A* path planning algorithm to plan an initial path in the grid map, and use the MinimumSnap motion planning algorithm to dynamically optimize the initial path. The dynamically feasible path will continue to be tested for safety, and if it is unsafe, it will continue to be optimized. S6: The detection drones exchange map information, undetected points of interest information, and task division results, and perform environmental detection along the planned safe path; The exploration drone and the inspection drone are both equipped with a gimbal camera to detect the surface of the object. The posture of the inspection drone and the gimbal camera are comprehensively controlled to balance the number of targets in the field of view and the inspection quality. The communication and integration module of the ground station coordinates the communication between the exploration drone and the detection drone, and timestamps the information. The information that does not participate in the logical calculation is based on the latest timestamp.

[0019] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A multi-UAV building detection method based on heuristic sampling and clustering, characterized by: The following steps are involved: S1. Use a grid map combined with an octree map to model the environment, taking into account search and computational efficiency. Based on the size and complexity of the map, the map resolution is independently selected to accelerate computation and query. S2, pre-processing the bounding box to accelerate the startup of the drone cluster and limit the working range of the drone cluster; S3: The exploration drone uses a lidar to model the scene. It visits each waypoint sequentially through a heuristic function to fully model the objects of interest in the environment. It also performs a rough inspection of the object surface during motion. The rough inspection results are provided to the detection drone as initial task division data. S4, multiple detection drones detect targets in their respective mission areas based on clustering division; S5. Use grid maps combined with octree maps to maximize search and computational efficiency to model and represent the environment for UAV motion path planning. S6. The detection drones exchange map information, information about undetected points of interest, and task division results, and perform environmental detection along the planned safe path.

2. The multi-UAV building detection method based on heuristic sampling and clustering as claimed in claim 1, characterized in that: In step S1, the environment includes the target building. Initially, all points in the map are set to be occupied. Through continuous scanning by the lidar, a safe space is calculated through line of sight. When the lidar illuminates a surface within the bounding box, the surface point set is supplemented and all surface points are set to an undetected state.

3. The multi-UAV building detection method based on heuristic sampling and clustering as claimed in claim 1, characterized in that: The specific process of preprocessing the bounding box in step S2 is as follows: S21. Sampling multiple points inside and near the bounding box according to a predetermined rule and marking them with features to accelerate subsequent calculations; wherein the feature markings include whether they are surface points, whether they are interior points, and whether they have been detected; S22. Sample multiple waypoints in the bounding box as candidate waypoints for exploring the drone.

4. The multi-UAV building detection method based on heuristic sampling and clustering as claimed in claim 1, characterized in that: In step S2, the bounding box is a three-dimensional frame containing the object of interest, and the target within the three-dimensional frame is modeled and detected by the drone cluster.

5. The multi-UAV building detection method based on heuristic sampling and clustering as claimed in claim 1, characterized in that: In step S3, for Waypoints , the heuristic function is: ,in, For the The heuristic value of each waypoint; For the The degree of unexplored and undetected areas near each waypoint; For the The distance between each waypoint and other drones of the same type; To go to The distance between waypoints; The distance to the aircraft that has not met the takeoff conditions is adopted; a reasonable distance strategy is adopted to maintain the obstacle. Based on the current map, the initial waypoint is offset. The offset direction is along the line connecting the current position to the center of gravity of the nearby obstacle, and the position is shifted forward or backward to the optimal observation distance; through communication, multiple exploration drones synchronize the explored waypoints with each other to avoid repeated exploration, and the maps are updated with each other during the communication process to provide an initial rough inspection result; wherein, the rough inspection result is a set of points of interest.

6. The multi-UAV building detection method based on heuristic sampling and clustering as claimed in claim 1, characterized in that: The specific processing process of step S4 is: S41. Based on the position of each detection drone, the initial rough inspection results are clustered based on distance, and each detection drone is assigned a portion of the mission area; The task allocation process is restarted after receiving new information greater than a threshold or after a period of time; S42. For each mission area, each detection UAV performs density-based clustering on known points of interest or surface points to be detected within the mission area to refine its own mission area. S43. Visit each task area in turn through a heuristic function to detect the target in the set of interest points.

7. The multi-UAV building detection method based on heuristic sampling and clustering as claimed in claim 1, characterized in that: In step S5, after the exploration drone or detection drone selects the next waypoint, it will use the A* path planning algorithm to plan an initial path in the grid map, and use the MinimumSnap motion planning algorithm to dynamically optimize the initial path; The safety of the dynamically feasible path will continue to be tested, and if it is unsafe, it will be continuously optimized.

8. The multi-UAV building detection method based on heuristic sampling and clustering as claimed in claim 1, characterized in that: The exploration drone and the detection drone are both equipped with a gimbal camera to detect the surface of objects. The posture of the detection drone and the gimbal camera are comprehensively controlled to balance the number of targets in the field of view and the detection quality.

9. The multi-UAV building detection method based on heuristic sampling and clustering as claimed in claim 1, characterized in that: The communication and integration module of the ground station coordinates the communication between the exploration drone and the detection drone, and timestamps the information. The information that does not participate in the logical calculation is based on the latest timestamp.

Images