Method for measuring three-dimensional route based on approaching photogrammetry of unmanned aerial vehicle

Abstract

The invention provides a method for measuring a three-dimensional route based on approaching photogrammetry of an unmanned aerial vehicle, which comprises the following steps of: S1, analyzing route planning methods based on a target initial model in different flight modes by utilizing route planning definition and two common methods for three-dimensional reconstruction of the unmanned aerial vehicle; according to concepts and processes close to photogrammetry, summarizing the requirements on route planning from the aspects of a planning method, a hardware platform and a route file format; s2, defining four types of aerial photography basic geometric objects used for selecting a to-be-reconstructed target in the initial model, and designing a surface target automatic extraction algorithm; and S3, according to constraint factors and a common method of three-dimensional path planning of the unmanned aerial vehicle, spatial obstacle avoidance analysis is performed through spatial environment modeling and collision detection, and a route optimization method including viewpoint adjustment and path search is designed on the basis of the spatial obstacle avoidance analysis. According to the method, aiming at the problem of unmanned aerial vehicle three-dimensional route planning in photogrammetry three-dimensional reconstruction work, photographing viewpoint planning meeting reconstruction requirements and unmanned aerial vehicle flight path planning considering various influence factors are researched.

Description

Technical Field

[0001] The invention belongs to the field of photogrammetry, and in particular relates to a method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle. Background Art

[0002] With the advancement of the construction of real-life 3D China, people's requirements for the refinement of 3D reconstruction are constantly increasing. Using multi-rotor drones equipped with high-resolution cameras for close-up photogrammetry is an emerging method for refined 3D reconstruction. This method uses a small and highly flexible multi-rotor drone to perform object-oriented photography of the target, obtain high-resolution images of the target surface at close range, and then achieve refined 3D reconstruction. However, for the route planning problem during image acquisition, unlike the two-dimensional routes or terrain-simulating routes in conventional vertical photogrammetry and oblique photogrammetry, close-up photogrammetry requires analysis of the initial model of the target and planning of a refined route in three-dimensional space. Therefore, in-depth research on the three-dimensional route planning method in close-up photogrammetry of drones is of great significance to the refined 3D reconstruction of real scenes. Summary of the invention

[0003] The present invention provides a method for three-dimensional route measurement based on unmanned aerial vehicle close-up photogrammetry. This method aims at the problem of three-dimensional route planning of unmanned aerial vehicle in three-dimensional reconstruction work based on close-up photogrammetry, and studies the photography viewpoint planning that meets the reconstruction requirements and the flight path planning of unmanned aerial vehicle considering multiple influencing factors.

[0004] In order to realize the above technical features, the object of the present invention is achieved as follows: a method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle comprises the following steps: S1, the definition and two common methods of route planning for 3D reconstruction using UAVs, and analysis of route planning methods based on the target initial model under different flight modes; based on the concept and process of photogrammetry, summarizes the requirements for route planning from the perspectives of planning methods, hardware platforms and route file formats; S2, define four types of aerial photography basic geometric objects used to select the target to be reconstructed in the initial model, and design an automatic extraction algorithm for surface targets; S3, according to the constraints and common methods of UAV three-dimensional path planning, spatial obstacle avoidance analysis is carried out through spatial environment modeling and collision detection, and on this basis, a route optimization method including viewpoint adjustment and path search is designed.

[0005] The specific process of the close-up photogrammetry work in S1 from coarse to fine is as follows: first, a rough three-dimensional model of the target area is obtained by conventional photography using a drone, then a close-up drone route is planned based on the initial model information and high-resolution images are collected, and finally, the images are three-dimensionally reconstructed to generate high-quality fine geographic information; The planning method in S1: Various geometric reconstruction constraints need to be considered comprehensively during route planning, including: sensor parameters, image resolution, intersection angle and overlap; In addition, in terms of path planning, route planning also requires spatial analysis of the target scene information, taking into account the spatial constraints of spatial occlusion, physical obstacles and no-fly zones; at the same time, it also needs to be combined with the UAV’s own flight control constraints, including: UAV power consumption, minimum distance limit and GNSS positioning error.

[0006] The hardware platform in S1: the photography gimbal used to carry the camera sensor in the UAV system can provide hardware support for this requirement. Its angle is continuously controllable and has high flexibility. It can adjust the photography angle of the camera sensor according to the planning results to achieve the target-oriented photography task by adjusting the angle in two degrees of freedom: horizontal rotation and lens pitch. The route file format in S1 adopts the KML file format. The route file recorded in KML contains the UAV yaw angle, gimbal pitch angle, photography action parameters and viewpoint longitude and latitude information.

[0007] The basic aerial photography geometric objects in S2 include: four basic aerial photography geometric objects used for target selection, namely point, line, surface and body; in the target selection operation, the defined basic geometric objects are used to divide and combine the surfaces of different modeling targets, so as to determine the surface points of the target object and discretize them to obtain the corresponding photographic points, and at the same time determine the photographic direction according to the surface normal vector of the target; any target entity can be expressed through these four types of basic geometric objects or their flexible combination, so as to achieve complete selection and coverage of targets in three-dimensional space, including local details of the targets that need to be restored for refined three-dimensional reconstruction.

[0008] The S2 mid-surface target grid segmentation and plane fitting method: (1) Calculate the flatness of all vertices of the 3D mesh in the k-ring neighborhood, and use the mean flatness of each vertex to represent the flatness of the corresponding facet; (2) Select the patch with the highest flatness as the seed, and use its k-ring neighborhood as the initial plane area to determine the reference plane through principal component analysis; (3) Check the k-ring neighborhood of the seed and merge the faces whose vertices are less than the threshold distance from the reference plane into the current plane area; (4) Repeat the above process until the entire three-dimensional grid is decomposed into various plane regions, the grid segmentation is completed, and the segmentation results of each plane region are obtained.

[0009] The S2 mid-surface target polygon extraction method: (1) Construct a Delaunay triangulation M for the input coplanar point set S; (2) Find the length of each edge and the set of adjacent triangles in the triangulated network M, and define the edge with only one adjacent triangle as the boundary edge of the point set; (3) Add all boundary edges whose length is greater than the length threshold lT to queue L. For an edge in L, delete T from the adjacent triangle sets of the other two edges of its only adjacent triangle T, and add the edge whose length is greater than lT to L. Repeat this step until L is empty. (4) Output all remaining boundary edges that are less than the threshold lT to represent the boundary shape of the point set.

[0010] The constraints of path planning in S3 are: (1) Safety constraints: When planning a path, the drone must always maintain a certain safe distance from other objects. In addition, for low-altitude areas, airspace restrictions such as no-fly zones or danger zones in addition to physical obstacles must also be considered. (2) Flight distance constraint: It is necessary to estimate the maximum range of the UAV while ensuring flight safety and plan an effective long-distance flight path within this constraint; (3) Turning radius constraint: Avoid right-angle turns when planning the flight path, and reduce the change in the flight angle of the drone; (4) Climb constraint: When planning the path, it is necessary to minimize the vertical altitude of the UAV to avoid unnecessary frequent altitude changes.

[0011] The path planning method in S3 can adopt a sampling-based algorithm, a node-based optimization algorithm, a mathematical model-based algorithm, a bionic algorithm or a fusion algorithm.

[0012] Spatial obstacle avoidance analysis in S3: In spatial obstacle avoidance analysis, the key to modeling the spatial environment with an octree structure is to determine the idle or occupied attributes of the voxel space. The octree modeling method divides the data in the three-dimensional space into blocks according to certain rules. Each division is based on the current node, and the space is recursively divided along the x, y, and z axes using three mutually perpendicular planes. Eight child nodes are generated each time until the size of the subspace reaches the set resolution size; for each voxel obtained by the segmentation, an attribute value is set to describe whether it is occupied. If the attribute is idle, it means that the place is free space and the drone can pass through it during flight. Otherwise, it means that the place is an obstacle and the path planning should avoid the position.

[0013] The route path optimization in S3: for any viewpoint, assuming that the drone needs to take pictures here, a collision detection is performed on the drone and the space environment model at this position. If no collision occurs, the current viewpoint is retained, otherwise the viewpoint needs to be adjusted. By adopting a viewpoint adjustment method that considers the constraints of the photographic incident angle while ensuring the photographic distance, in an actual environment, the closer to the ground, the greater the possibility of obstacles. Based on this, the search is preferentially performed in the vertical direction to obtain a photographic position that can ensure flight safety. Then, the photographic direction at the photographic position is determined according to the spatial position relationship between the target and the drone, and the adjustment of this viewpoint is completed.

[0014] Beneficial effects of the present invention: 1. The present invention proposes a photography design method driven by geometric entities, which generates viewpoint positions and photography directions that meet reconstruction requirements based on target surface information.

[0015] 2. Based on spatial analysis, the present invention designs a UAV path planning method that takes obstacle avoidance and distance constraints into consideration. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0017] Figure 1 To be close to the photogrammetry flow chart.

[0018] Figure 2 Flowchart of the three-dimensional route planning method based on photogrammetry.

[0019] Figure 3 Automatically extract flow charts for area objects.

[0020] Figure 4 Schematic diagram of the octree structure.

[0021] Figure 5 Flowchart for modeling octree space.

[0022] Figure 6 Adjust the diagram for the viewpoint. DETAILED DESCRIPTION

[0023] The embodiments of the present invention are further described below in conjunction with the accompanying drawings.

[0024] Embodiment 1: The present invention provides a method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle, which is specifically implemented by executing the following steps: S1. Define the route planning for 3D reconstruction using drones and two common methods, and analyze the route planning methods based on the target initial model under different flight modes. Based on the concept and process of photogrammetry, summarize the requirements for route planning from the perspectives of planning methods, hardware platforms, and route file formats; S2. Define four types of aerial photography basic geometric objects for selecting the target to be reconstructed in the initial model, and design an automatic extraction algorithm for surface targets; S3. According to the constraints and common methods of UAV three-dimensional path planning, spatial obstacle avoidance analysis is performed through spatial environment modeling and collision detection, and on this basis, a route optimization method including viewpoint adjustment and path search is designed; Furthermore, the specific process of the close-up photogrammetry work in S1 from coarse to fine is as follows: first, a rough three-dimensional model of the target area is obtained by conventional photography using a drone; then, a close-flying drone route is planned based on the initial model information and high-resolution images are collected; finally, the images are three-dimensionally reconstructed to generate high-quality fine geographic information.

[0025] Furthermore, the planning method in S1: various geometric reconstruction constraints need to be considered during route planning, such as sensor parameters, image resolution, intersection angle, and overlap, etc. In addition, in terms of path planning, route planning also requires spatial analysis of target scene information, considering spatial constraints such as spatial occlusion, physical obstacles, and no-fly zones, and also needs to be combined with the UAV's own flight control constraints, such as UAV power consumption, minimum distance limit, and GNSS positioning error, etc.

[0026] Furthermore, the hardware platform in S1: the photography gimbal used to carry the camera sensor in the UAV system can provide hardware support for this requirement. Its angle is continuously controllable and has high flexibility. The photography angle of the camera sensor can be adjusted according to the planning results to achieve target-oriented photography tasks, generally for the angle adjustment in two degrees of freedom: horizontal rotation and lens pitch.

[0027] Furthermore, the route file format in S1: KML (Keyhole Markup Language) file format is a coding specification for describing and saving geographic information, which uses the syntax format of extensible markup language to describe geographic spatial data. KML file format is a commonly used route file format. The route file recorded in KML contains information such as the drone's yaw angle, gimbal pitch angle, photography action parameters, and viewpoint latitude and longitude.

[0028] Furthermore, the basic aerial photography geometric objects in S2 are: four common basic aerial photography geometric objects used for target selection, namely point, line, surface and body. In the target selection operation, the defined basic geometric objects can be used to divide and combine the surfaces of different modeling targets, so as to determine the surface points of the target object and discretize them to obtain the corresponding photographic points, and at the same time determine the photographic direction according to the surface normal vector of the target. Any target entity can be expressed through these four types of basic geometric objects or their flexible combination to achieve complete selection and coverage of targets in three-dimensional space, including local details of the target that need to be restored for refined three-dimensional reconstruction.

[0029] Furthermore, the surface target grid segmentation and plane fitting method in S2: (1) Calculate the flatness of all vertices of the 3D mesh in the k-ring neighborhood, and use the mean flatness of each vertex to represent the flatness of the corresponding facet; (2) Select the patch with the highest flatness as the seed, and use its k-ring neighborhood as the initial plane area to determine the reference plane through principal component analysis (PCA). (3) Check the k-ring neighborhood of the seed and merge the faces whose vertices are less than the threshold distance from the reference plane into the current plane area; (4) Repeat the above process until the entire three-dimensional grid is decomposed into various plane regions, the grid segmentation is completed, and the segmentation results of each plane region are obtained.

[0030] Furthermore, the surface target polygon extraction method in S2 is: (1) Construct a Delaunay triangulation M for the input coplanar point set S; (2) Find the length of each edge and the set of adjacent triangles in the triangulated network M, and define the edge with only one adjacent triangle as the boundary edge of the point set; (3) Add all boundary edges whose length is greater than the length threshold lT to queue L. For an edge in L, delete T from the adjacent triangle sets of the other two edges of its only adjacent triangle T, and add the edge whose length is greater than lT to L. Repeat this step until L is empty. (4) Output all remaining boundary edges that are less than the threshold lT to represent the boundary shape of the point set.

[0031] Furthermore, the constraints of the path planning in S3 are: (1) Safety constraints: Ensuring the flight safety of drones is one of the key issues in path planning. The higher flight speed of drones reduces the pressure in the surrounding space, increasing the possibility of collision with other objects in the space. Therefore, when planning paths, drones must always maintain a certain safe distance from other objects. In addition, for low-altitude areas, airspace restrictions such as no-fly zones or danger zones in addition to physical obstacles need to be considered; (2) Flight distance constraint: The endurance of a light rotor UAV is usually very limited. Therefore, it is necessary to estimate the maximum range of the UAV while ensuring flight safety and plan an effective flight path that is as long as possible within this constraint. (3) Turning radius constraint: Unlike fixed-wing drones, multi-rotor drones rely on the rotors on each arm to provide flight power and have the ability to make right-angle turns. However, considering flight energy consumption, right-angle turns are usually not used when planning flight paths, and the flight angle changes of the drone should be minimized. (4) Climb constraint: Although multi-rotor UAVs have the ability to climb vertically, the power consumption during vertical climbing is often higher than that during horizontal flight, which results in a reduction in the UAV's flight time. Therefore, when planning the path, it is necessary to minimize unnecessary and frequent altitude changes of the UAV in the vertical direction.

[0032] Furthermore, the commonly used methods for path planning in S3 are: (1) Sampling-based algorithms: This type of method samples the environment in nodes or other forms based on known environmental spatial information, and then maps the environment into a network graph or performs a random search to connect the nodes to find the best path. Depending on whether a final path is generated directly, this type of algorithm can be divided into two subcategories: active algorithms and passive algorithms. At present, a large number of algorithms based on different core ideas and their improved methods have been proposed, which can be mainly divided into the Rapidly-Exploring Random Trees (RRT) series, artificial potential field series, Voronoi diagram series and probabilistic roadmaps (PRM) series. This type of method is easy to implement and has a simple structure. It is suitable for static and online planning conditions; (2) Node-based optimization algorithm: This type of algorithm processes the nodes in the graph or map results obtained by preprocessing the environment space. This type of algorithm is a special form of dynamic programming. It defines the cost function, converts the distance between nodes into calculation weights, and then searches to determine a path with the lowest cost as the global optimal path. Commonly used methods include Dijkstra algorithm, A* algorithm, D* algorithm, and Harmony Search (HS). This type of method can be combined with other methods to achieve global optimization and is suitable for online work; (3) Algorithms based on mathematical models: This type of method describes various constraints in mathematical form and models the spatial environment, and then constrains the cost function with inequalities or equations to obtain the optimal solution. This type of method includes linear programming, optimal control, etc., which usually requires high computer computing power and is therefore generally suitable for some offline work; (4) Bionic algorithms: Bionic algorithms are based on human learning behavior and imitate biological behavior to solve problems, eliminating the process of building complex environmental models and proposing a powerful search method to stably converge to the target. This type of method can be divided into neural network (NN) algorithms and evolutionary algorithms (EA). This type of algorithm is a heuristic algorithm that can handle complex unstructured constraints and other non-deterministic problems well. The solution process of this type of algorithm usually requires a long iteration time, so it is only suitable for offline work; (5) Fusion algorithm: Considering the complexity of the actual situation, the current three-dimensional path planning method tends to achieve the goal of global optimization and minimum cost by integrating the advantages of multiple algorithms. This type of algorithm can be divided into two categories according to the fusion method. One category integrates multiple algorithms and searches for the optimal path at the same time; the other category combines different algorithms in a designed order to meet the task requirements. In some specific cases, better planning results can be obtained by integrating several relatively simple algorithms, which can be used for online implementation.

[0033] Furthermore, the spatial obstacle avoidance analysis in S3: In the spatial obstacle avoidance analysis, the key to modeling the spatial environment with the octree structure is to determine the idle or occupied attributes of the voxel space. The octree is a commonly used data structure model, that is, a tree structure with eight child nodes, each node representing a cube in a three-dimensional space, also called a voxel. The octree modeling method divides the data in the three-dimensional space into blocks according to certain rules. Each division is based on the current node, and the space is recursively divided along the x, y, and z axes using three mutually perpendicular planes, generating eight child nodes each time until the size of the subspace reaches the set resolution size. For each voxel obtained by segmentation, an attribute value is set to describe whether it is occupied. If the attribute is idle, it means that the place is free space and the drone can pass through it during flight. Otherwise, it means that the place is an obstacle and the path planning should avoid the position.

[0034] Furthermore, the route path optimization in S3: for any viewpoint, assuming that the drone needs to shoot here, the drone and the space environment model are subjected to collision detection at this position. If no collision occurs, the current viewpoint is retained, otherwise the viewpoint needs to be adjusted. This method adopts a viewpoint adjustment method that considers the constraints of the photographic incident angle while ensuring the photographic distance. In the actual environment, the closer to the ground, the greater the possibility of obstacles. Therefore, this method preferentially searches in the vertical direction to obtain a photographic position that can ensure flight safety. Then, the photographic direction at the photographic position is determined according to the spatial position relationship between the target and the drone to complete the adjustment of this viewpoint.

Claims

1. A method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle, characterized in that: The steps include: S1, the definition and two common methods of route planning for 3D reconstruction using UAVs, and analysis of route planning methods based on the target initial model under different flight modes; based on the concept and process of photogrammetry, summarizes the requirements for route planning from the perspectives of planning methods, hardware platforms and route file formats; S2, define four types of aerial photography basic geometric objects used to select the target to be reconstructed in the initial model, and design an automatic extraction algorithm for surface targets; S3, according to the constraints and common methods of UAV three-dimensional path planning, spatial obstacle avoidance analysis is carried out through spatial environment modeling and collision detection, and on this basis, a route optimization method including viewpoint adjustment and path search is designed.

2. According to claim 1, a method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle is characterized in that: The specific process of the close-up photogrammetry work in S1 from coarse to fine is as follows: first, a rough three-dimensional model of the target area is obtained by conventional photography using a drone, then a close-up drone route is planned based on the initial model information and high-resolution images are collected, and finally, the images are three-dimensionally reconstructed to generate high-quality fine geographic information; The planning method in S1: Various geometric reconstruction constraints need to be considered comprehensively during route planning, including: sensor parameters, image resolution, intersection angle and overlap; In addition, in terms of path planning, route planning also requires spatial analysis of the target scene information, taking into account the spatial constraints of spatial occlusion, physical obstacles and no-fly zones; at the same time, it also needs to be combined with the UAV’s own flight control constraints, including: UAV power consumption, minimum distance limit and GNSS positioning error.

3. According to claim 2, a method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle is characterized in that: The hardware platform in S1: the photography gimbal used to carry the camera sensor in the UAV system can provide hardware support for this requirement. Its angle is continuously controllable and has high flexibility. It can adjust the photography angle of the camera sensor according to the planning results to achieve the target-oriented photography task by adjusting the angle in two degrees of freedom: horizontal rotation and lens pitch. The route file format in S1 adopts the KML file format. The route file recorded in KML contains the UAV yaw angle, gimbal pitch angle, photography action parameters and viewpoint longitude and latitude information.

4. The method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle according to claim 1, characterized in that: The basic aerial photography geometric objects in S2 include: four basic aerial photography geometric objects used for target selection, namely point, line, surface and body; in the target selection operation, the defined basic geometric objects are used to divide and combine the surfaces of different modeling targets, so as to determine the surface points of the target object and discretize them to obtain the corresponding photographic points, and at the same time determine the photographic direction according to the surface normal vector of the target; any target entity can be expressed through these four types of basic geometric objects or their flexible combination, so as to achieve complete selection and coverage of targets in three-dimensional space, including local details of the targets that need to be restored for refined three-dimensional reconstruction.

5. The method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle according to claim 4, characterized in that: The S2 mid-surface target grid segmentation and plane fitting method: (1) Calculate the flatness of all vertices of the 3D mesh in the k-ring neighborhood, and use the mean flatness of each vertex to represent the flatness of the corresponding facet; (2) Select the patch with the highest flatness as the seed, and use its k-ring neighborhood as the initial plane area to determine the reference plane through principal component analysis; (3) Check the k-ring neighborhood of the seed and merge the faces whose vertices are less than the threshold distance from the reference plane into the current plane area; (4) Repeat the above process until the entire three-dimensional grid is decomposed into various plane regions, the grid segmentation is completed, and the segmentation results of each plane region are obtained.

6. The method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle according to claim 5, characterized in that: The S2 mid-surface target polygon extraction method: (1) Construct a Delaunay triangulation M for the input coplanar point set S; (2) Find the length of each edge and the set of adjacent triangles in the triangulated network M, and define the edge with only one adjacent triangle as the boundary edge of the point set; (3) Add all boundary edges whose length is greater than the length threshold lT to queue L. For an edge in L, delete T from the adjacent triangle sets of the other two edges of its only adjacent triangle T, and add the edge whose length is greater than lT to L. Repeat this step until L is empty. (4) Output all remaining boundary edges that are less than the threshold lT to represent the boundary shape of the point set.

7. The method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle according to claim 1, characterized in that: The constraints of path planning in S3 are: (1) Safety constraints: When planning a path, the drone must always maintain a certain safe distance from other objects. In addition, for low-altitude areas, airspace restrictions such as no-fly zones or danger zones in addition to physical obstacles must also be considered. (2) Flight distance constraint: It is necessary to estimate the maximum range of the UAV while ensuring flight safety and plan an effective long-distance flight path within this constraint; (3) Turning radius constraint: Avoid right-angle turns when planning the flight path, and reduce the change in the flight angle of the drone; (4) Climb constraint: When planning the path, it is necessary to minimize the vertical altitude of the UAV to avoid unnecessary frequent altitude changes.

8. The method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle according to claim 1, characterized in that: The path planning method in S3 can adopt a sampling-based algorithm, a node-based optimization algorithm, a mathematical model-based algorithm, a bionic algorithm or a fusion algorithm.

9. The method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle according to claim 8, characterized in that: Spatial obstacle avoidance analysis in S3: In spatial obstacle avoidance analysis, the key to modeling the spatial environment with an octree structure is to determine the idle or occupied attributes of the voxel space. The octree modeling method divides the data in the three-dimensional space into blocks according to certain rules. Each division is based on the current node, and the space is recursively divided along the x, y, and z axes using three mutually perpendicular planes. Eight child nodes are generated each time until the size of the subspace reaches the set resolution size; for each voxel obtained by the segmentation, an attribute value is set to describe whether it is occupied. If the attribute is idle, it means that the place is free space and the drone can pass through it during flight. Otherwise, it means that the place is an obstacle and the path planning should avoid the position.

10. The method for measuring a three-dimensional route based on close-up photography of an unmanned aerial vehicle according to claim 9, characterized in that: The route path optimization in S3: for any viewpoint, assuming that the drone needs to take pictures here, a collision detection is performed on the drone and the space environment model at this position. If no collision occurs, the current viewpoint is retained, otherwise the viewpoint needs to be adjusted. By adopting a viewpoint adjustment method that considers the constraints of the photographic incident angle while ensuring the photographic distance, in an actual environment, the closer to the ground, the greater the possibility of obstacles. Based on this, the search is preferentially performed in the vertical direction to obtain a photographic position that can ensure flight safety. Then, the photographic direction at the photographic position is determined according to the spatial position relationship between the target and the drone, and the adjustment of this viewpoint is completed.

Images