A large complex curved surface full coverage three-dimensional path planning method

Abstract

The application discloses a large complex curved surface full coverage three-dimensional path planning method, mainly aiming at unmanned aerial vehicle path planning for shape defect detection of an aircraft surface, and the method comprises the following steps: firstly, a view cone principle is adopted to offset down-sampling of aircraft point clouds to construct an unmanned aerial vehicle observation view point and an observation pose; further, an improved A* search algorithm is utilized to construct a connecting edge in a neighborhood of the observation point; finally, an ant colony algorithm is utilized to search a shortest full coverage path according to graph constraints, so that unmanned aerial vehicle coverage path planning based on the aircraft point clouds is realized; and finally, a MiniSnap method is adopted to perform trajectory optimization on the path points, so that the path points meet unmanned aerial vehicle dynamics; wherein the improved A* search algorithm is mainly optimized from two aspects of an expansion node direction and construction of an evaluation function. Finally, the method realizes an automatic generation method from an aircraft model to a full traversal path, solves problems of complex manual trajectory planning, non-optimal planned path and low efficiency, and has the characteristics of automatic generation and optimal trajectory.

Description

Technical Field

[0001] This invention belongs to the field of industrial inspection and relates to a three-dimensional path planning method for full coverage of large and complex curved surfaces. Background Technology

[0002] Routine aircraft inspection is a crucial means of ensuring safe flight. Currently, 90% of aircraft skin inspections are conducted manually by visual inspection or by hand-held inspection equipment. These methods suffer from visual fatigue, reduced efficiency, and the time-consuming and labor-intensive process of climbing ladders to inspect the high tail section of the aircraft. Therefore, it is proposed to use drones equipped with cameras to collect images of the outer surface and further utilize deep learning for inspection. The coverage path planning of the drone is a critical component that directly affects the final drone scanning efficiency and obstacle avoidance safety.

[0003] The outer surface of an aircraft is a large and complex curved surface. Routine aircraft inspection can be abstracted as a coverage path planning problem on a large and complex curved surface. Coverage path planning is to determine an obstacle-avoidable path that can pass through all regions of interest. Currently, the coverage path planning problem can be roughly divided into two types: stochastic methods and unit decomposition heuristic function methods. The idea behind stochastic methods is similar to the diffusion phenomenon of molecules undergoing Brownian motion—given a sufficiently long running time, all regions can be covered according to a certain pattern. The unit decomposition heuristic function method decomposes the space to be traversed into multiple cells. The traversal of each cell adopts a fixed traversal trajectory, similar to a zigzag trajectory. Then, the traversal of multiple cells is transformed into a traveling salesman problem, ultimately achieving the overall path node traversal.

[0004] According to the search, Guo's "Path planning algorithm for sweeping robot full traversal cleaning area" is based on the idea of ​​cell decomposition. First, the region is divided as a whole, then local path planning is carried out, and finally a global path planning algorithm is used to connect multiple cells to form the final path, thus realizing the construction of a full traversal path.

[0005] Searching revealed that Emile, in its work "Overage path planning with targeted viewpoint sampling for robotic free form surface inspection," transformed the problem of selecting observation points covering freeform surfaces into a joint optimization problem that ensures the same viewpoint can see as many observation primitives as possible and that the covered observation primitives of different viewpoints are as different as possible. By using a weighted approach, under the constraints of robot kinematics and obstacle collision-free conditions, the optimal solution is sought, and the optimization effect is ultimately verified to be significant.

[0006] According to the search, Dai Jiajia proposed an improved WaveFront coverage path planning algorithm in her research on aircraft appearance inspection methods based on machine vision technology. This algorithm solves the local optimum problem caused by the same loss value in different directions based on the constraint strategy of adjacent viewpoints, and designs a three-dimensional spatial direction evaluation function to realize the coverage path planning of UAVs.

[0007] In summary, the acquisition of overlay images of the UAV's outer surface can be abstracted as a set relation problem, involving two sets (S, Q). S is a set of geometric primitives, including points, voxels, and patches, while Q is the robot's shape space. The goal is to find a subset P of Q, and the set SM, which contains the mappings from each element of subset P to S. Since the aircraft's outer surface is a large and complex curved surface, the path for full coverage of the outer surface normal needs to balance scanning efficiency and collision avoidance. Using a stochastic method to solve the overlay path planning problem cannot guarantee timeliness in three-dimensional space. While the element decomposition method can achieve overall traversal, it requires consideration of the relationship between the shooting angle and the position of the acquisition viewpoint, and the viewpoint acquisition route must meet obstacle avoidance requirements. Summary of the Invention

[0008] The purpose of this invention is to provide a three-dimensional path planning method for full coverage of large and complex curved surfaces. It is mainly applied to the use of UAVs to achieve traversal detection of the outer surface of aircraft, ensuring obstacle avoidance, detection effectiveness, and planning timeliness.

[0009] To achieve the above objectives, the specific design of this invention is as follows:

[0010] A method for full-coverage 3D path planning of large and complex curved surfaces, the flowchart of which is as follows: Figure 1 As shown, it includes the following steps:

[0011] Step 1: Constructing the observation point location: Based on the aircraft point cloud model and the view frustum model, the point cloud is downsampled to obtain the central viewpoint. The point cloud is segmented using the RANSAC method to obtain the normals. The central viewpoint is then offset to obtain the observation point location.

[0012] Step 2: Search graph construction: Based on the central viewpoint position obtained during the observation point location construction process, the improved A* algorithm is used to construct the connection paths between points in the neighborhood and construct the cost matrix.

[0013] Step 3: Drone traversal path generation: Based on the directed connected graph constructed from the search graph, the ant colony algorithm is used to solve the problem of full traversal path planning and optimization.

[0014] Step 4: Drone traversal trajectory generation: The MiniSnap method is used to smooth the constructed traversal path, ensuring that its position, velocity, and acceleration are smooth, and passing through the corresponding path nodes.

[0015] Furthermore, the first step, the construction of the observation point location, is as follows: First, based on the complete model file, the point cloud location is collected to obtain the dense point cloud map of the object to be detected in the original file; second, the central viewpoint of the observation is obtained through the point cloud downsampling method, and the point cloud normal direction of the central viewpoint is obtained by using the RANSAC method based on the points around the central viewpoint; finally, the central viewpoint is offset along the point cloud normal direction to obtain the observation viewpoint collected by the full traversal, and its normal direction is the observation direction.

[0016] Furthermore, the second step, graph construction, is as follows: First, an observation viewpoint is used to find its nearest neighbor viewpoints; second, an obstacle-avoiding path is constructed between the observation viewpoint and its nearest neighbor viewpoints using an improved A* algorithm, while obtaining the cost of the obstacle-avoiding path as the edge length of the graph construction, and finally obtaining a directed connected graph.

[0017] Furthermore, in the second step, the improved A* algorithm is as follows: First, the direction of the A* expansion node is restricted, so that it only expands along the surrounding points within the direction angle of the target node; second, the cost function for the expansion node of A* is modified to adopt a combined distance, which can ensure that the distance function is closer to the expansion node method.

[0018] Furthermore, in the third step, the directed connection graph is traversed without repetition using the ant colony algorithm. Then, the obstacle-avoidable paths of the corresponding connection points generated in claim 5 are read sequentially according to the order of traversal of the nodes, and finally the full traversal path is obtained.

[0019] Furthermore, the fourth step is specifically operated as follows: the UAV traversal trajectory is generated based on the full traversal path generated in the third step. The MiniSnap algorithm is used to minimize the fifth derivative to generate a smooth full traversal trajectory with position, velocity, and acceleration. This path meets the UAV dynamics requirements.

[0020] The beneficial effects of this invention are:

[0021] This method realizes an automatic generation method from aircraft model to full traversal path, which solves the problems of complexity, non-optimal planning path and low efficiency of manual trajectory planning. It has the characteristics of automatic generation and optimal trajectory. Attached Figure Description

[0022] Figure 1 This is a flowchart of the three-dimensional path planning method for full coverage of large and complex curved surfaces according to the present invention.

[0023] Figure 2 This is a flowchart illustrating the construction process of observation points in this invention.

[0024] Figure 3This is a rasterized point cloud image of the present invention.

[0025] Figure 4 This is a diagram showing the observation viewpoint results of this invention;

[0026] Figure 5 This invention relates to a graph construction method;

[0027] Figure 6 This is a flowchart of the collision detection process of the present invention;

[0028] Figure 7 This is the connection diagram result of the present invention;

[0029] Figure 8 This is a flowchart of the ant colony algorithm of the present invention;

[0030] Figure 9 This is the iterative optimal path diagram for this invention;

[0031] Figure 10 This is a graph showing the shortest path convergence result of the present invention. Detailed Implementation

[0032] The technical solution of this patent will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the examples of this patent, and not all of them. Based on the embodiments of this patent, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this patent.

[0033] The embodiments of this patent are described in detail below. Examples of these embodiments are shown 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 are only used to explain this patent, and should not be construed as limiting this patent.

[0034] like Figure 2 As shown, the observation point construction is based on an aircraft point cloud model. This model is used to obtain a dense aircraft point cloud through point cloud sampling. The central viewpoint is then determined through downsampling. The downsampling size is chosen based on the frustum principle, selecting the field-of-view plane size calculated for the corresponding camera model. The downsampling principle involves rasterizing the point cloud, as shown... Figure 3 As shown, by calculating the size of the centroid corresponding to each dimension, a representative centroid point within the cell is selected as the central viewpoint. The RANSAC method is used to obtain the normals, finding the point cloud around the center point, fitting a plane to the current point cloud plane, and finally calculating the plane's normal vector to obtain the current node's normal. Finally, to obtain the pose that satisfies the camera's observation, the center point is translated along the normal direction to obtain the corresponding offset center point, i.e., the observation point; the corresponding normal direction is the observation's line of sight. The final observation viewpoint result is shown in the image below. Figure 4 As shown.

[0035] like Figure 5 The diagram shows the flowchart of the graph construction method. The graph is the foundation for the full traversal algorithm. Graph elements include nodes and corresponding directed edges. The node elements are the center offset points mentioned above, i.e., the observation viewpoints. A directed edge is an edge that points to two observation nodes. In this patent, due to application requirements, the actual meaning of this edge is to connect two nodes without collisions. Considering that constructing connecting edges between nodes that are far apart is meaningless and will be discarded in the final traversal, this patent uses nearest neighbor points within a radius of 10m to construct connecting edges. The method for constructing collision-free paths uses an improved A* algorithm. The final constructed connected graph is shown below. Figure 7 As shown.

[0036] like Figure 6 The diagram illustrates how collision detection is performed using the A* algorithm. Considering that traditional obstacle avoidance detection divides the space into grids and expands grid nodes, this patent uses a direct collision detection method based on the point cloud for node expansion. Specifically, a KD-tree is constructed along the dimensional direction of the aircraft point cloud, and then the dense aircraft point cloud is divided into a tree structure. The collision detection of expanded nodes is then performed using a KD-tree search, which has a computational complexity of log n, thus saving computation time.

[0037] like Figure 8 As shown, based on the already constructed graph connections, all nodes are traversed, and the connecting paths between corresponding nodes are linked to construct a complete full traversal path planning method. The traversal of all nodes uses the ant colony algorithm, a swarm intelligence algorithm that simulates ants foraging for food. Ants tend to take the shortest path when searching for food, releasing pheromones during their movement. Thus, in a given unit of time, the pheromone residue on the shorter path is greater than that on the longer path, leading all ants to eventually take the shortest path. Specifically, this includes the ants choosing their next viewpoint and updating pheromones.

[0038] The probability construction and pheromone update are key aspects of this application. The probability construction is shown in the following formula:

[0039]

[0040] Where i and j represent the start and end points respectively, η ij =1 / d ij Visibility, expressed here as the reciprocal of the distance between the two observation nodes, τ ij (t) represents the pheromone concentration between the two observation points at time t, allowed.k Let represent the set of unvisited observation points and expandable observation points, where α and β represent the weights between pheromone and visibility, respectively.

[0041] The update of pheromones is as follows:

[0042]

[0043] Where m represents the number of "ants" and ρ represents the pheromone evaporation rate, with 0 < ρ ≤ 1. Let pheromones be the pheromones left by the k-th ant along path ij. The final optimal path obtained through the full traversal is as follows: Figure 10 As shown in the figure, its convergence effect is as follows: Figure 9 As shown. To further ensure the continuity of each node and satisfy human-machine dynamics, i.e., the continuity between position, velocity, and acceleration, MiniSnap is used to smooth and optimize the obtained convergent path nodes to obtain the UAV's flight trajectory.

[0044] In summary, this invention proposes a complete technical route for UAVs to fully traverse the outer surface of an aircraft, consisting of observation viewpoint construction, graph construction, and graph traversal. Ultimately, using an aircraft model as data support, the process begins with camera selection and image plane calculation. Based on the image plane size, a reasonable observation viewpoint is constructed. An improved A* algorithm is used to construct the graph, and finally, an ant colony algorithm is employed to traverse the graph. Analysis of the final simulation results shows that the traversal effect is achieved, and the traversal cost can be significantly reduced.

[0045] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for full-coverage three-dimensional path planning of large and complex curved surfaces, characterized in that, Includes the following steps: Step 1: Constructing the observation point location: Based on the aircraft point cloud model and the view frustum model, the point cloud is downsampled to obtain the central viewpoint. The point cloud is segmented using the RANSAC method to obtain the normals. The central viewpoint is then offset to obtain the observation point location. The first step, observation point location construction, is as follows: First, based on the complete model file, point cloud locations are collected to obtain the dense point cloud map of the object to be detected in the original file; second, the central viewpoint of observation is obtained through point cloud downsampling method, and the point cloud normal direction of the central viewpoint is obtained based on the points around the central viewpoint using the RANSAC method; finally, the central viewpoint is offset along the point cloud normal direction to obtain the observation viewpoint collected by the full traversal, and its normal direction is the observation direction; Step 2: Search Graph Construction: Based on the location of the observation point, a search graph is constructed to obtain the locations of other observation points in the neighborhood, using an improved A* algorithm. The algorithm constructs connection paths between observation points in the neighborhood and builds a cost matrix; In the second step, the improved A The algorithm is as follows: First, for A... The direction of the expanded node is restricted, so that it only expands along the peripheral points within the direction angle of the target node; secondly, for A The valuation cost function for the expansion nodes is modified to use a combined distance, which ensures that the distance function is closer to the expansion node method. Step 3: Drone traversal path generation: Based on the directed connected graph constructed from the search graph, the ant colony algorithm is used to solve the problem of full traversal path planning and optimization. Step 4: Drone traversal trajectory generation: The Mini Snap method is used to smooth the constructed traversal path, ensuring that its position, velocity, and acceleration are smooth, and passing through the path nodes of interest.

2. The method for full-coverage three-dimensional path planning of large and complex curved surfaces as described in claim 1, characterized in that, The second step, the search graph construction, is as follows: First, using the observation viewpoint, find the nearest neighbor viewpoints; second, connect the observation viewpoint and its nearest neighbor viewpoints using an improved A... The algorithm constructs an obstacle-avoidable path between two points and obtains the cost of the obstacle-avoidance path, which is used as the length of the edge in the graph construction, and finally obtains a directed connected graph.

3. The method for full-coverage three-dimensional path planning of large and complex curved surfaces as described in claim 2, characterized in that, In the third step, the directed connection graph is traversed without repetition using the ant colony algorithm. Then, the obstacle avoidance paths of the corresponding connection points generated in claim 2 are read sequentially according to the order of traversal of the nodes, and finally the full traversal path is obtained.

4. The method for full-coverage three-dimensional path planning of large and complex curved surfaces as described in claim 1, characterized in that, In the fourth step, the Mini Snap method is used to minimize the fifth derivative to generate a smooth full traversal trajectory of position, velocity, and acceleration, which satisfies the dynamic requirements of the UAV.

Images