Unmanned aerial vehicle exploration system based on incremental boundary information and layered exploration planning

Abstract

The invention discloses an unmanned aerial vehicle exploration system based on incremental boundary information and layered exploration planning, and the system remarkably improves the exploration efficiency and response speed of an unmanned aerial vehicle in a complex environment through an incremental boundary information module and a layered exploration planning module. The incremental boundary information module can efficiently process new information and quickly adapt to environmental changes, so that the unmanned aerial vehicle can complete exploration tasks at the speed 3-8 times that of an existing method. The layered exploration planning module finds an efficient global path covering all boundary clusters by converting an exploration task into a variant of a travel salesman problem, and avoids low efficiency of a local greedy strategy. Meanwhile, the hierarchical exploration planning module further refines local viewpoints on the basis of global path planning, selects an optimal viewpoint combination, and enhances the flexibility and efficiency of exploration.

Description

Technical Field

[0001] The present invention relates to the field of autonomous exploration technology of multi-rotor unmanned aerial vehicles, and in particular to an unmanned aerial vehicle exploration system based on incremental boundary information and hierarchical exploration planning. Background Art

[0002] Due to their high maneuverability, rapid deployment, and relatively low cost, multi-rotor drones have been widely used in search and rescue, agricultural monitoring, and industrial monitoring. Drones use sensors to perceive unknown environments in real time, build maps, and intelligently plan a safe, efficient, and dynamically constrained exploration path to maximize information collection and coverage. This process requires drones to quickly adapt to environmental changes and dynamically optimize exploration strategies.

[0003] At present, UAV autonomous exploration technology mainly relies on advanced perception and planning algorithms to achieve efficient exploration of unknown environments. These technologies usually include environmental perception, path planning, and adaptive processing of dynamic environments. However, these mainstream technologies have exposed some significant deficiencies in practical applications: 1. Limitations of exploration strategies: Many existing methods adopt greedy strategies (e.g., A. Bircher, M. Kamel, K. Alexis, H. Oleynikova, and R. Siegwart, “Receding horizon” next-best-view” planner for 3 d exploration,” in Proc. IEEE Int. Conf. Robot. Autom., 2016, pp. 1462–1468), such as maximizing immediate information gain or navigating to the nearest unknown area. This strategy ignores global optimality, resulting in low overall exploration efficiency. This is because the greedy strategy only considers the local optimal solution, but does not consider the efficiency and effect of global coverage.

[0004] 2. Conservatism of motion planning: In order to ensure both information gain and safety in unknown environments, most methods generate relatively conservative motion plans (e.g., M. Selin, M. Tiger, D. Duberg, F. Heintz, and P.Jensfelt, “Efficient autonomous exploration planning of large-scale 3Denvironments,” IEEE Robot. Automat. Lett., vol. 4, no. 2, pp. 1699–1706, Apr. 2019). This limits the full use of the UAV’s dynamic capabilities, resulting in a reduction in exploration speed and an inability to fully utilize the UAV’s high-speed maneuverability advantage.

[0005] 3. Computational overhead and decision frequency: Some methods cannot respond quickly to environmental changes due to high computational overhead (e.g., L. Schmid, M. Pantic, R. Khanna, L. Ott, R. Siegwart, and J. Nieto, “Anefficient sampling-based method for online informative path planning in unknown environments,” IEEE Robot. Automat. Lett., vol. 5, no. 2, pp. 1500–1507, Apr. 2020). In complex environments, making decisions quickly is crucial to improving exploration efficiency. However, existing methods often take a long time to process large amounts of data and perform complex calculations, which affects the frequency of decision-making and response speed.

[0006] 4. Adaptability to environmental changes: Existing methods usually fail to fully consider the dynamic changes of the environment during planning, such as the sudden appearance of obstacles or changes in the target position. This results in the drone being unable to adjust its exploration strategy in time when the environment changes, thus affecting the completion of the exploration mission. Summary of the invention

[0007] In view of the shortcomings of the prior art, the present invention aims to provide a UAV exploration system based on incremental boundary information and hierarchical exploration planning.

[0008] In order to achieve the above object, the present invention adopts the following technical solution: A UAV exploration system based on incremental boundary information and hierarchical exploration planning, including an incremental boundary information module and a hierarchical exploration planning module; The incremental boundary information module is used to store and update the environmental information required by the drone during the exploration process; when a boundary cluster is created, the incremental boundary information module generates a corresponding incremental boundary area information unit, and stores the calculated information of all grids belonging to the boundary cluster and the average position of all grids and the axis-aligned bounding box of the boundary cluster in the incremental boundary area information unit; the incremental boundary information module generates candidate viewpoints around the boundary cluster for each boundary cluster and calculates a bidirectional linked list of the boundary cluster and saves it in the corresponding incremental boundary area information unit, and the bidirectional linked list contains the connection cost between the boundary cluster to which it belongs and all other boundary clusters; when a boundary cluster is created, the incremental boundary information module generates a set of candidate viewpoints for the boundary cluster, and these candidate viewpoints cover the boundary cluster; after generating the candidate viewpoints, the viewpoints with a coverage range higher than the preset coverage range threshold are retained, and the retained viewpoints are sorted in descending order according to the size of the coverage range, and the viewpoint with the highest coverage range is the first one in the sorting; The hierarchical exploration and planning module includes a global exploration path planning module, a local viewpoint refinement module, and a shortest time B-Spline trajectory generation module; The global exploration path planning module is used to calculate the optimal open-loop travel route for the UAV starting from the current viewpoint and visiting the viewpoint with the highest coverage of all boundary clusters; The local viewpoint refinement module is used to find the optimal local travel route on the truncated segment of the optimal open-loop travel route obtained by the global exploration path planning module according to the multiple viewpoints retained by each boundary cluster; The optimal local path obtained by the local viewpoint refinement module consists of discrete path points. The shortest time B-Spline trajectory generation module is used to use the discrete path points as the initial reference path, generate a smooth, safe and dynamically feasible trajectory through fitting, and minimize the total trajectory time by optimizing parameters.

[0009] Furthermore, whenever the drone updates the map through sensor measurements, the incremental boundary information module records the axis-aligned bounding box of the updated area. ; In the update region, if the axis-aligned bounding box of the old bounding cluster and If they intersect, the incremental boundary information module removes the corresponding old boundary clusters.

[0010] Furthermore, after removing the outdated boundary clusters, the incremental boundary information module searches for new boundaries and clusters them; for the boundary clusters obtained by clustering, if the number of grids contained is less than the set grid number threshold, the boundary cluster is ignored, and for the large-sized boundary cluster whose size exceeds the set size threshold, the incremental boundary information module takes the longest direction of the large-sized boundary cluster as the first principal axis direction, and then divides the large-sized boundary cluster into two smaller-sized boundary clusters along the first principal axis through principal component analysis, and if the maximum eigenvalue of the segmented boundary cluster exceeds the preset eigenvalue threshold, the segmented boundary cluster continues to be segmented until the maximum eigenvalue of the segmented boundary cluster does not exceed the preset eigenvalue threshold; the maximum eigenvalue of the boundary cluster represents the maximum variance of the size of the boundary cluster in the direction of the first principal axis.

[0011] Furthermore, the candidate viewpoints of each boundary cluster are obtained by uniformly sampling in a cylindrical coordinate system with the center of the boundary cluster as the origin.

[0012] Furthermore, for each candidate viewpoint, the corresponding drone yaw angle is the direction of the maximum range boundary cluster covered by the visible area of ​​the drone's sensor.

[0013] Furthermore, the incremental boundary information module controls the number of retained viewpoints.

[0014] Furthermore, after deleting the old boundary cluster, the incremental boundary information module deletes the connection cost items about the old boundary cluster in the bidirectional linked lists corresponding to all remaining boundary clusters, and then when adding a new boundary cluster, calculates the connection cost of the new boundary cluster to each other boundary cluster and inserts it into the bidirectional linked list generated for the new boundary cluster.

[0015] Furthermore, the connection cost is calculated as follows: Indicates that the cluster is at the boundary Viewpoint in and boundary clusters Viewpoint in The lower bound of time when moving between is: in, Representative Viewpoint The three-dimensional position coordinates of Representative Viewpoint The three-dimensional position coordinates of Represents the path found by the path search algorithm (A* algorithm) on the voxel grid map and The collision-free path between Represents the drone at the viewpoint The yaw angle at Represents the drone at the viewpoint The yaw angle at Representing from the point of view To Viewpoint The minimum angle the drone needs to rotate. and Respectively represent the maximum limits of the speed and yaw rate of the drone; Select the boundary clusters separately and The viewpoint with the highest coverage in , and the boundary cluster and The estimated connection cost between .

[0016] Furthermore, the global exploration path planning module calculates the cost matrix , find the optimal closed-loop travel route, and then get the open-loop travel route with equal cost; assuming that there are a total of Boundary clusters, cost matrix Corresponding to one +1-dimensional square matrix, the main part of which is composed of the connection cost between each pair of boundary clusters block, the cost matrix is ​​calculated as follows: in, and Represents the boundary clusters and The first-ranked viewpoint is retained. The first row and column of the current viewpoint and The boundary clusters are associated with and Respectively represent the position and yaw angle of the drone in the current viewpoint; from the current viewpoint Initially, the connection cost to the kth boundary cluster is evaluated as follows: in, represents the motion consistency cost; Represents the motion consistency cost The weight represents the penalty for large changes in flight direction and is an adjustable parameter. Motion consistency cost The calculation method is as follows: in, Indicates the current speed of the drone; By setting the connection cost between other boundary clusters and the current viewpoint to 0, we have: Thus, returning to the current viewpoint in any closed-loop travel route does not add extra cost, and each closed-loop travel route always contains an open-loop path with the same cost. Therefore, the optimal open-loop travel route is obtained by finding the optimal closed-loop travel route and retrieving its open-loop travel route with equivalent cost.

[0017] Furthermore, the process of obtaining the optimal local path by the local viewpoint refinement module is as follows: first, find the boundary clusters with similar positions. Specifically, in the circle with the current viewpoint position of the drone as the center and the radius as The boundary clusters corresponding to the viewpoints within the viewpoint refinement range are the boundary clusters with similar positions; let is the considered boundary cluster, is the number of boundary clusters considered within the viewpoint refinement range; is the total number of retained viewpoints of the considered boundary clusters and the current viewpoint position Create graph nodes; then each graph node is connected to the graph nodes corresponding to the viewpoints retained by the adjacent boundary clusters, forming a directed acyclic graph that captures the changes in possible truncated travel, and then the Dijkstra algorithm is used to search for the optimal local path.

[0018] The beneficial effects of the present invention are: 1. The system of the present invention significantly improves the exploration efficiency and response speed of drones in complex environments through the incremental boundary information module and the hierarchical exploration planning module. The incremental boundary information module can efficiently process new information and quickly adapt to environmental changes, allowing drones to immediately respond to newly discovered unknown areas, allowing drones to complete exploration tasks at a faster speed and increase decision-making frequency.

[0019] 2. In the present invention, the hierarchical exploration planning module finds an efficient global path covering all boundary clusters by converting the exploration task into a variant of the traveling salesman problem (TSP), avoiding the inefficiency of the local greedy strategy. At the same time, the hierarchical exploration planning module further refines the local viewpoints on the basis of global path planning, selects the optimal viewpoint combination, and enhances the flexibility and efficiency of exploration. The hierarchical exploration planning module also uses the shortest time B-Spline trajectory planning to optimize the trajectory smoothness and total flight time, while considering the dynamic limitations of the drone. This enables the drone to execute a smooth and dynamic flight trajectory, fully utilize its flight performance, and improve flight speed and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 An example diagram of removing and updating boundaries by an incremental boundary information module in an embodiment of the present invention; Figure 2 An example diagram of generating candidate viewpoints by an incremental boundary information module in an embodiment of the present invention; Figure 3 This is an example diagram of the results obtained by the hierarchical exploration planning module in an embodiment of the present invention after global exploration path planning, local viewpoint refinement, and shortest time B-Spline trajectory generation; Figure 4 This is an example diagram of the generation result of local viewpoint refinement in an embodiment of the present invention; Figure 5 A schematic diagram showing the results of applying the technical solution of the present invention to indoor experiments; Figure 6 A schematic diagram showing the results of applying the technical solution of the present invention to outdoor experiments. DETAILED DESCRIPTION

[0021] The present invention will be further described below in conjunction with the accompanying drawings. It should be noted that this embodiment is based on the technical solution and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to this embodiment.

[0022] This embodiment provides a UAV exploration system based on incremental boundary information and hierarchical exploration planning, including an incremental boundary information module and a hierarchical exploration planning module; 1. Incremental Boundary Information Module The incremental boundary information module is used to store and update the key environmental information required by the drone during the exploration process. When a boundary cluster is created, the incremental boundary information module generates a corresponding incremental boundary area information unit and stores the calculated information of all grids belonging to the boundary cluster, the average position of all grids, and the axis-aligned bounding box (AABB) of the boundary cluster in the incremental boundary area information unit. It should be noted that the axis-aligned bounding box of the boundary cluster helps to speed up the detection of boundary changes.

[0023] Figure 1 In the figure, the upper left side shows that the incremental boundary information module detects the old boundary cluster, the upper right side shows that the incremental boundary information module removes the old boundary cluster, and the lower left side shows that the incremental boundary information module detects the new boundary cluster. Through principal component analysis, the large boundary cluster is divided into two smaller boundary clusters, such as Figure 1 as shown below on the right.

[0024] In this embodiment, in order to serve the exploration planning, the incremental boundary information module generates candidate viewpoints around the boundary cluster for each boundary cluster and calculates the bidirectional linked list of the boundary cluster and saves it in the corresponding incremental boundary area information unit. The bidirectional linked list contains the connection cost between the boundary cluster to which it belongs and all other boundary clusters.

[0025] In this embodiment, in order to efficiently detect and cluster boundaries, each time the drone updates the map through sensor measurements, the incremental boundary information module records the axis-aligned bounding box of the updated area. In the update region, if the axis-aligned bounding box of the old bounding cluster and If the clusters intersect, the incremental boundary information module removes the corresponding old boundary clusters. This process can quickly exclude most of the unaffected clusters and significantly reduce the number of precise checks.

[0026] After removing the outdated boundary clusters, the incremental boundary information module searches for new boundaries and clusters them using the region growing algorithm. For the boundary clusters obtained by clustering, if the number of grids contained is less than the set grid number threshold, the boundary cluster is ignored. For large-sized boundary clusters whose size exceeds the set size threshold, the incremental boundary information module uses the longest direction of the large-sized boundary cluster as the direction of the first principal axis, and divides the large-sized boundary cluster into two smaller-sized boundary clusters along the first principal axis through principal component analysis (PCA). If the maximum eigenvalue of the segmented boundary cluster exceeds the preset eigenvalue threshold, the segmented boundary cluster continues to be segmented until the maximum eigenvalue of the segmented boundary cluster does not exceed the preset eigenvalue threshold. In PCA, the covariance matrix of the data is used to describe the degree of change of the data in each direction. The eigenvalues ​​of the covariance matrix correspond to the variance of the data in the direction of the principal components, and the eigenvectors indicate the directions of these principal components. In this embodiment, the maximum eigenvalue of the boundary cluster represents the maximum variance of the size of the boundary cluster in the direction of the first principal axis. If the maximum eigenvalue exceeds the set threshold, it means that the boundary cluster is too long in the direction of the first principal axis and needs to be further segmented to ensure that the size of all boundary clusters is small enough to facilitate subsequent processing.

[0027] In this embodiment, when a boundary cluster is created, the incremental boundary information module generates a set of candidate viewpoints for the boundary cluster, which cover the boundary cluster. The candidate viewpoints are obtained by uniformly sampling in a cylindrical coordinate system with the center of the boundary cluster as the origin. Figure 2 An example of generating candidate viewpoints.

[0028] For each candidate viewpoint, the corresponding drone yaw angle is the direction of the maximum range boundary cluster covered by the drone's sensor's visible area (that is, at this yaw angle, the drone's sensor can see the most points in the boundary cluster). After generating candidate viewpoints, viewpoints with coverage higher than the preset coverage threshold are retained, and the retained viewpoints are sorted in descending order by coverage size, and the first one is the viewpoint with the highest coverage. It should be noted that the incremental boundary information module controls the number of retained viewpoints to make local viewpoint refinement feasible.

[0029] In this embodiment, the connection cost between boundary clusters is updated in an incremental manner. After deleting the old boundary cluster, the incremental boundary information module deletes the connection cost item about the old boundary cluster in the bidirectional linked list corresponding to all remaining boundary clusters, and then when adding a new boundary cluster, the connection cost of the new boundary cluster to each other boundary cluster is calculated and inserted into the bidirectional linked list generated for the new boundary cluster. This incremental update scheme makes the update time of the connection cost linearly related to the number of new boundary clusters, thereby achieving efficient update. In this way, each incremental boundary area information unit provides strong support for the rapid exploration of drones in complex environments.

[0030] More specifically, the connection cost is calculated as follows: set up Indicates that the cluster is at the boundary Viewpoint in and boundary clusters Viewpoint in The lower bound of time when moving between is: in, Representative Viewpoint The three-dimensional position coordinates of Representative Viewpoint The three-dimensional position coordinates of Represents the path found by the path search algorithm (A* algorithm) on the voxel grid map and The collision-free path between Represents the drone at the viewpoint The yaw angle at Represents the drone at the viewpoint The yaw angle at Representing from the point of view To Viewpoint The minimum angle the drone needs to rotate. and Respectively represent the maximum limits of the speed and yaw rate of the drone; Select the boundary clusters separately and The viewpoint with the highest coverage in , and the boundary cluster and The estimated connection cost between .

[0031] 2. Hierarchical Exploration Planning Module The hierarchical exploration and planning module includes a global exploration path planning module, a local viewpoint refinement module and a shortest time B-Spline trajectory generation module.

[0032] In this embodiment, the exploration task of the drone is regarded as a variant of the Traveling Salesman Problem (TSP), and the open-loop travel route of the drone starting from the current viewpoint and visiting the viewpoint with the highest coverage of all boundary clusters is calculated. The global exploration path planning module calculates the cost matrix , find the optimal closed-loop travel route, and then get the open-loop travel route with equal cost. Assume that there are a total of Boundary clusters, cost matrix Corresponding to one +1-dimensional square matrix, the main part of which is composed of the connection cost between each pair of boundary clusters block, the cost matrix is ​​calculated as follows: in, and Represents the boundary clusters and The first-ranked viewpoint is retained. The first row and column of the current viewpoint and The boundary clusters are associated with and Represents the position and yaw angle of the drone in the current viewpoint. Initially, the connection cost to the kth boundary cluster can be evaluated as follows: in, represents the motion consistency cost; Represents the motion consistency cost The weight represents the penalty for large changes in flight direction. It is an adjustable parameter that users can adjust according to actual conditions.

[0033] Motion consistency cost The calculation method is as follows: in, Represents the current speed of the drone. In some cases, multiple travel routes have similar lower bounds on time, so back-and-forth maneuvers may be generated in consecutive plans, slowing down progress. To eliminate this inconsistency, it penalizes large changes in flight direction.

[0034] The drone exploration in this embodiment is different from the standard TSP, whose solution is a closed-loop travel route. However, this embodiment simplifies it to an asymmetric traveling salesman problem (ATSP) by setting the connection cost between other boundary clusters and the current viewpoint to 0: In this way, returning to the current viewpoint in any closed-loop travel route does not add additional cost, and each closed-loop travel route always contains an open-loop path with the same cost. Therefore, the optimal open-loop travel route can be obtained by finding the optimal closed-loop travel route and retrieving its equivalent cost open-loop travel route.

[0035] The local viewpoint refinement module is used to find the optimal local travel route on the truncated segment of the optimal open-loop travel route obtained by the global exploration path planning module based on the multiple viewpoints retained by each boundary cluster. This stage improves the exploration efficiency by considering more viewpoint combinations. The specific process is as follows: First, find the boundary clusters with similar positions. Specifically, with the current viewpoint of the drone as the center and the radius as The boundary clusters corresponding to the viewpoints within the viewpoint refinement range of 5m (artificially set in this embodiment) are the boundary clusters with similar positions. is the considered boundary cluster, is the number of boundary clusters considered within the viewpoint refinement range. is the number of all retained viewpoints within the considered boundary clusters and the current viewpoint position Create graph nodes. Each graph node is then connected to the graph nodes corresponding to the viewpoints retained by the adjacent boundary clusters, forming a directed acyclic graph that captures the possible changes in truncated travel. Then use the Dijkstra algorithm to search for the optimal local path. The results are shown in the example below. Figure 4 shown.

[0036] The minimum time B-Spline trajectory generation module is used to generate continuous trajectories for smooth navigation. The optimal local path obtained by the local viewpoint refinement module consists of discrete path points (viewpoints), and its path may have discontinuous corners or non-smooth transitions. In order to consider the dynamic constraints of the UAV (such as maximum speed, acceleration, angular velocity, etc.), the minimum time B-Spline trajectory generation module uses the discrete path points as the initial reference path, generates a smooth, safe and dynamically feasible trajectory by fitting, and minimizes the total trajectory time by optimizing parameters, so that the UAV can fully utilize its dynamic capabilities and improve the efficiency of the UAV's global exploration. The trajectory planning considers the differential flatness characteristics of the UAV (e.g., D.Mellinger and V. Kumar, “Minimum snap trajectory generation and control forquadrotors,” in Proc. IEEE Int. Conf. Robot. Autom., May 2011, pp. 2520–2525.), and uses elastic band smoothness cost, total trajectory time, safety and dynamic feasibility penalties, etc. to formulate the optimization problem.

[0037] Figure 3 Shown is an example of the processing results of the hierarchical exploration planning module.

[0038] Through extensive testing in indoor and outdoor environments, the system of this embodiment has proven its effectiveness in complex real-world scenarios, solving the problems of low efficiency, slow response and insufficient planning in the prior art. Figure 5 This is the indoor experimental result diagram (the left side is the real environment, the right side is the experimental result). Figure 6 This is the outdoor experimental result diagram (the left side is the real environment, and the right side is the experimental result).

[0039] For those skilled in the art, various corresponding changes and modifications can be made according to the above technical solutions and concepts, and all of these changes and modifications should be included in the protection scope of the claims of the present invention.

Claims

1. A UAV exploration system based on incremental boundary information and hierarchical exploration planning, characterized by: Includes incremental boundary information module and hierarchical exploration planning module; The incremental boundary information module is used to store and update the environmental information required by the drone during the exploration process; when a boundary cluster is created, the incremental boundary information module generates a corresponding incremental boundary area information unit, and stores the calculated information of all grids belonging to the boundary cluster and the average position of all grids and the axis-aligned bounding box of the boundary cluster in the incremental boundary area information unit; the incremental boundary information module generates candidate viewpoints around the boundary cluster for each boundary cluster and calculates a bidirectional linked list of the boundary cluster and saves it in the corresponding incremental boundary area information unit, and the bidirectional linked list contains the connection cost between the boundary cluster to which it belongs and all other boundary clusters; when a boundary cluster is created, the incremental boundary information module generates a set of candidate viewpoints for the boundary cluster, and these candidate viewpoints cover the boundary cluster; after generating the candidate viewpoints, the viewpoints with a coverage range higher than the preset coverage range threshold are retained, and the retained viewpoints are sorted in descending order according to the size of the coverage range, and the viewpoint with the highest coverage range is the first one in the sorting; The hierarchical exploration and planning module includes a global exploration path planning module, a local viewpoint refinement module, and a shortest time B-Spline trajectory generation module; The global exploration path planning module is used to calculate the optimal open-loop travel route for the UAV starting from the current viewpoint and visiting the viewpoint with the highest coverage of all boundary clusters; The local viewpoint refinement module is used to find the optimal local travel route on the truncated segment of the optimal open-loop travel route obtained by the global exploration path planning module according to the multiple viewpoints retained by each boundary cluster; The optimal local path obtained by the local viewpoint refinement module consists of discrete path points. The shortest time B-Spline trajectory generation module is used to use the discrete path points as the initial reference path, generate a smooth, safe and dynamically feasible trajectory through fitting, and minimize the total trajectory time by optimizing parameters.

2. The system according to claim 1, characterized in that Whenever the drone updates the map via sensor measurements, the incremental bounding information module records the axis-aligned bounding box of the updated area. ; In the update region, if the axis-aligned bounding box of the old bounding cluster and If they intersect, the incremental boundary information module removes the corresponding old boundary clusters.

3. The system according to claim 2, characterized in that After removing the outdated boundary clusters, the incremental boundary information module searches for new boundaries and clusters them; for the boundary clusters obtained by clustering, if the number of grids contained is less than the set grid number threshold, the boundary cluster is ignored; for large-sized boundary clusters whose size exceeds the set size threshold, the incremental boundary information module uses the longest direction of the large-sized boundary cluster as the first principal axis direction, and then divides the large-sized boundary cluster into two smaller-sized boundary clusters along the first principal axis through principal component analysis. If the maximum eigenvalue of the segmented boundary cluster exceeds the preset eigenvalue threshold, the segmented boundary cluster continues to be segmented until the maximum eigenvalue of the segmented boundary cluster does not exceed the preset eigenvalue threshold; the maximum eigenvalue of the boundary cluster represents the maximum variance of the size of the boundary cluster in the direction of the first principal axis.

4. The system according to claim 1, characterized in that The candidate viewpoints of each boundary cluster are obtained by uniformly sampling in a cylindrical coordinate system with the center of the boundary cluster as the origin.

5. The system according to claim 1 or 4, characterized in that: For each candidate viewpoint, its corresponding drone yaw angle is the direction of the maximum range boundary cluster covered by the visible area of ​​the drone's sensor.

6. The system according to claim 1, characterized in that The incremental boundary information module controls the number of retained viewpoints.

7. The system according to claim 3, characterized in that After deleting the old boundary cluster, the incremental boundary information module deletes the connection cost items about the old boundary cluster in the bidirectional linked lists corresponding to all remaining boundary clusters. Then, when adding a new boundary cluster, it calculates the connection cost of the new boundary cluster to other boundary clusters and inserts it into the bidirectional linked list generated for the new boundary cluster.

8. The system according to claim 1, characterized in that The connection cost is calculated as follows: set up Indicates that the cluster is at the boundary Viewpoint in and boundary clusters Viewpoint in The lower bound of time when moving between is: in, Representative Viewpoint The three-dimensional position coordinates of Representative Viewpoint The three-dimensional position coordinates of Represents the path found by the path search algorithm (A* algorithm) on the voxel grid map and The collision-free path between Represents the drone at the viewpoint The yaw angle at Represents the drone at the viewpoint The yaw angle at Representing from the point of view To Viewpoint The minimum angle the drone needs to rotate. and Respectively represent the maximum limits of the speed and yaw rate of the UAV; Select the boundary clusters separately and The viewpoint with the highest coverage in , and the boundary cluster and The estimated connection cost between .

9. The system according to claim 1, characterized in that The global exploration path planning module calculates the cost matrix , find the optimal closed-loop travel route, and then get the open-loop travel route with equal cost; assuming that there are a total of Find the optimal closed-loop travel route, and then get the open-loop travel route with equal cost; assuming there are a total of Corresponding to one +1-dimensional square matrix, the main part of which is composed of the connection cost between each pair of boundary clusters block, the cost matrix is ​​calculated as follows: in, and Represents the boundary clusters and The first-ranked viewpoint is retained. The first row and column of the current viewpoint and The boundary clusters are associated with and Respectively represent the position and yaw angle of the drone in the current viewpoint; from the current viewpoint Initially, the connection cost to the kth boundary cluster is evaluated as follows: in, represents the motion consistency cost; Represents the motion consistency cost The weight represents the penalty for large changes in flight direction and is an adjustable parameter. Motion consistency cost The calculation method is as follows: in, Indicates the current speed of the drone; By setting the connection cost between other boundary clusters and the current viewpoint to 0, we have: Thus, returning to the current viewpoint in any closed-loop travel route does not add extra cost, and each closed-loop travel route always contains an open-loop path with the same cost. Therefore, the optimal open-loop travel route is obtained by finding the optimal closed-loop travel route and retrieving its open-loop travel route with equivalent cost.

10. The system according to claim 9, characterized in that The process of obtaining the optimal local path by the local viewpoint refinement module is as follows: First, find the boundary clusters with similar positions. Specifically, take the current viewpoint of the drone as the center and the radius as The boundary clusters corresponding to the viewpoints within the viewpoint refinement range are the boundary clusters with similar positions; let is the considered boundary cluster, is the number of boundary clusters considered within the viewpoint refinement range; is the total number of retained viewpoints of the considered boundary clusters and the current viewpoint position Create graph nodes; then each graph node is connected to the graph nodes corresponding to the viewpoints retained by the adjacent boundary clusters, forming a directed acyclic graph that captures the changes in possible truncated travel, and then the Dijkstra algorithm is used to search for the optimal local path.

Images