Unmanned aerial vehicle inspection optimization method and system for nature reserve

Abstract

The present application relates to the technical field of data processing, and particularly relates to a natural reserve unmanned aerial vehicle inspection optimization method, which solves the technical problem of unreasonable inspection route planning of an unmanned aerial vehicle in inspecting a natural reserve in the prior art. The method comprises the following steps: dividing the natural reserve into a plurality of grids, calculating the static obstacle density of the non-obstacle grids in the plurality of grids based on the point cloud data of the static obstacles of the natural reserve, and calculating the dynamic obstacle density of the non-obstacle grids based on the image data of the dynamic obstacles of the natural reserve; determining the comprehensive obstacle factor of each non-obstacle grid based on the static obstacle density and the dynamic obstacle density; selecting the initial point of the grey wolf algorithm based on the static obstacle density, and performing iterative calculation of the grey wolf algorithm with the minimum weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path as the optimization target to determine the inspection path.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a method and system for optimizing unmanned aerial vehicle (UAV) inspection of a nature reserve. BACKGROUND

[0002] Currently, an UAV is usually used to inspect and monitor a nature reserve. In the process of using the UAV to conduct the inspection, a grey wolf algorithm can be used to plan an inspection path for the UAV. However, when the grey wolf algorithm is used to plan the inspection path, the initial points of the grey wolf algorithm are usually selected to be more uniformly distributed. However, the uniformly distributed initial points of the grey wolf algorithm can cause the selected initial points of the grey wolf algorithm to not match the actual target distribution of the nature reserve, and the dynamic targets (such as bird targets) are not considered in the iteration process of the algorithm. Therefore, the inspection path planned based on the grey wolf algorithm has the technical problem of unreasonable inspection route. SUMMARY

[0003] In order to solve the technical problem of unreasonable inspection route planning for the UAV when inspecting the nature reserve in the related art, the purpose of the present application is to provide a method and system for optimizing UAV inspection of a nature reserve, and the technical solutions adopted are as follows:

[0004] The nature reserve is divided into a plurality of grids, the static obstacle density of the obstacle-free grids in the plurality of grids is calculated based on the point cloud data of the static obstacles of the nature reserve, and the dynamic obstacle density of the obstacle-free grids is calculated based on the image data of the dynamic obstacles of the nature reserve; wherein the static obstacle density is used to represent the distribution density of the static obstacles around the grid, and the dynamic obstacle density is used to represent the distribution density of the dynamic obstacles around the grid;

[0005] The comprehensive obstacle factor of each obstacle-free grid is determined based on the static obstacle density and the dynamic obstacle density;

[0006] The initial points of the grey wolf algorithm are selected based on the static obstacle density, and the iteration calculation of the grey wolf algorithm is performed with the minimum weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path as the optimization target to determine the inspection path.

[0007] In one possible implementation, the method comprises:

[0008] The obstacle grids and the obstacle-free grids in the plurality of grids are determined based on the point cloud data of the static obstacles of the nature reserve;

[0009] The obstacle grids are clustered based on a clustering algorithm to obtain a clustering area of the obstacle grids;

[0010] The first area of the obstacle grids in each clustering area is calculated, and the second area of the minimum circumscribed rectangle of each clustering area is calculated;

[0011] divide the first area of each clustering area by the second area to obtain a static obstacle density of each clustering area;

[0012] diffuse the static obstacle density of each clustering area to the unobstacle grid within the clustering area to obtain a static obstacle density of the unobstacle grid.

[0013] In a possible implementation, the method comprises:

[0014] calculate a first diffusion space of a target clustering area based on an area of the target clustering area; wherein the target clustering area is one of the clustering areas, and the first diffusion space is positively correlated with the area of the target clustering area;

[0015] input the first diffusion space of the target clustering area and a spatial distance between the first unobstacle grid and the target clustering area into a first Gaussian distribution function to determine a first Gaussian function value of the first unobstacle grid; the first unobstacle grid is one of the unobstacle grids within the target clustering area;

[0016] normalize a product of the static obstacle density of the target clustering area and the first Gaussian function value of the first unobstacle grid to obtain a static obstacle density of the first unobstacle grid.

[0017] In a possible implementation, the first Gaussian distribution function comprises a first exponential term and a first constant term, the first constant term is an inverse of a product of a square of the first diffusion space and 2π, and an exponent value of the first exponential term is a negative value of a quotient of the spatial distance and a square of 2 times the first diffusion space.

[0018] In a possible implementation, the method comprises:

[0019] collect a plurality of image frames of the nature reserve at preset time intervals, and detect a dynamic target in each image frame based on a dynamic target detection model;

[0020] determine an actual motion trajectory of the dynamic target based on a dynamic tracking algorithm, and process the actual motion trajectory based on a Kalman filtering algorithm to predict a predicted motion trajectory of the dynamic target in a future time period;

[0021] map the actual motion trajectory and the predicted motion trajectory to a plurality of grids, and determine, from the plurality of grids, a dynamic obstacle grid comprising the actual motion trajectory and / or the predicted motion trajectory, and an adjacent grid having a distance less than a preset distance from the dynamic obstacle grid;

[0022] calculate a dynamic obstacle density of a second unobstacle grid based on a total area of the dynamic obstacle grid and the adjacent grid, and a distance between the dynamic target and the second unobstacle grid in the adjacent grid.

[0023] In a possible implementation, the method comprises:

[0024] calculating a second diffusion space based on the dynamic obstacle grid and the total area of the adjacent grid, the second diffusion space being positively correlated with the area of the dynamic obstacle grid and the adjacent grid;

[0025] bringing the second diffusion space and the distance between the dynamic target and the second obstacle-free grid into a second Gaussian distribution function to determine a second Gaussian function value of the second obstacle-free grid;

[0026] normalizing the product of the second Gaussian function value and a dynamic obstacle area indication function of the second obstacle-free grid to obtain a dynamic obstacle density of the second obstacle-free grid.

[0027] In a possible implementation, the second Gaussian distribution function comprises a second exponential term and a second constant term, the second constant term being the inverse of the product of the square of the second diffusion space and 2π, and the exponent value of the second exponential term being the negative value of the quotient of the distance between the dynamic target and the second obstacle-free grid and twice the square of the second diffusion space.

[0028] In a possible implementation, the method comprises:

[0029] setting a set of inspection path points covering the nature reserve;

[0030] selecting an inspection path starting point from the set of inspection path points and calculating the minimum Euclidean distance between each obstacle-free grid in the adjacent grid of the grid to which the inspection path starting point belongs and the path points in the set of inspection path points;

[0031] based on the minimum Euclidean distance, calculating a minimum penalty distance, the minimum penalty distance being inversely proportional to the minimum Euclidean distance;

[0032] determining the product of the minimum penalty distance and the static obstacle density of each obstacle-free grid, and selecting the obstacle-free grid with the maximum product as the initial point of the grey wolf algorithm.

[0033] In a possible implementation, the method comprises:

[0034] determining a target function of the grey wolf algorithm, the target function being used to minimize the weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path;

[0035] setting the size of the grey wolf population and the maximum number of iterations, and performing iterative calculation of the grey wolf algorithm based on the target function to select an optimal grey wolf; during the iterative calculation, the positions of all the grey wolves are dynamically updated based on gradient approximation values, the size of the grey wolf population is set to 50, and the maximum number of iterations is set to 100;

[0036] taking the grid to which the optimal grey wolf belongs as the grid in the inspection path.

[0037] The application also provides a natural reserve unmanned aerial vehicle inspection optimization system, which comprises an obstacle factor calculation module and a path planning module.

[0038] The obstacle factor calculation module is used for dividing the natural reserve into a plurality of grids, calculating the static obstacle density of the non-obstacle grid in the plurality of grids based on the point cloud data of the static obstacles of the natural reserve, and calculating the dynamic obstacle density of the non-obstacle grid based on the image data of the dynamic obstacles of the natural reserve; wherein the static obstacle density is used for representing the distribution density of the static obstacles around the grid, and the dynamic obstacle density is used for representing the distribution density of the dynamic obstacles around the grid.

[0039] The obstacle factor calculation module is further used for determining the comprehensive obstacle factor of each non-obstacle grid based on the static obstacle density and the dynamic obstacle density.

[0040] The path planning module is used for selecting the initial point of the grey wolf algorithm based on the static obstacle density, and performing the iterative calculation of the grey wolf algorithm with the minimum weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path as the optimization target to determine the inspection path.

[0041] The application has the following beneficial effects: in the process of planning the inspection path of the unmanned aerial vehicle based on the grey wolf algorithm, the initial point of the grey wolf algorithm is selected based on the static obstacle density in each grid, the high-density static target area can be monitored, the inspection efficiency is improved, and the inspection route is more reasonable; in addition, when the grey wolf algorithm is iterated, the iterative calculation of the grey wolf algorithm is performed with the minimum weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path as the optimization target, so that the inspection route is the shortest and the probability of collision between the unmanned aerial vehicle and the obstacle is the lowest. Compared with the technical scheme in the prior art in which the selected initial point is uniformly distributed and the dynamic target is not considered, the natural reserve unmanned aerial vehicle inspection optimization method provided by the application can make the inspection path focus on monitoring the high-density area, and the planned inspection path can ensure the economy and safety of the inspection, so that the planned inspection route is more reasonable. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the application or the prior art, the drawings needed in the following embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0043] Figure 1 A natural reserve unmanned aerial vehicle inspection optimization method flow chart provided by an embodiment of the applicationFigure 1 ;

[0044] Figure 2 A natural reserve unmanned aerial vehicle inspection optimization method flowchart provided by an embodiment of the present application Figure 2 ;

[0045] Figure 3 A natural reserve unmanned aerial vehicle inspection optimization method flowchart provided by an embodiment of the present application Figure 3 ;

[0046] Figure 4 A natural reserve unmanned aerial vehicle inspection optimization method flowchart provided by an embodiment of the present application Figure 4 ;

[0047] Figure 5 A natural reserve unmanned aerial vehicle inspection optimization method flowchart provided by an embodiment of the present application Figure 5 ;

[0048] Figure 6 A natural reserve unmanned aerial vehicle inspection optimization method flowchart provided by an embodiment of the present application Figure 6 ;

[0049] Figure 7 A natural reserve unmanned aerial vehicle inspection optimization system schematic diagram provided by an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined purposes, the following describes in detail the specific embodiments, structures, features and effects of the natural reserve unmanned aerial vehicle inspection optimization method and system according to the present application, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0052] The specific scheme of the natural reserve unmanned aerial vehicle inspection optimization method and system provided by the present application is specifically described below in combination with the accompanying drawings.

[0053] Please refer to Figure 1 , which shows a flowchart of the natural reserve unmanned aerial vehicle inspection optimization method provided by an embodiment of the present application. The method includes the following steps:

[0054] Step 101, divide the nature reserve into a plurality of grids.

[0055] In a possible implementation, the area to be inspected by the unmanned aerial vehicle is divided into grid maps of the same size by using the grid method, and the grid maps include a plurality of grids. Further, the grids are divided into obstacle grids (which can also be referred to as non-traversable areas) and non-obstacle grids (which can also be referred to as traversable areas) based on whether there is a static obstacle in the grid. When planning the inspection route of the unmanned aerial vehicle, the unmanned aerial vehicle is preferentially planned to traverse in the non-obstacle grid. Optionally, the non-obstacle grid is marked as 0 to represent a traversable area, and the obstacle grid is marked as 1 to represent a non-traversable area.

[0056] Optionally, before planning the inspection route of the unmanned aerial vehicle, first, the point cloud data and image data of the nature reserve are collected by using an image collection device such as a light detection and ranging (LiDAR) and a high-definition camera. The point cloud data includes point cloud data of static targets such as trees, ground, and the like, and the image data includes image data of dynamic targets such as owls and bees. The point cloud data and the image data are corresponded to the grid map to obtain the point cloud data and the image data in each grid.

[0057] Step 102, calculate the static obstacle density of the non-obstacle grid in the plurality of grids based on the point cloud data of the static obstacle of the nature reserve.

[0058] The static obstacle density is used to represent the distribution density of the static obstacle around the grid.

[0059] In the embodiment of the present application, the static obstacle in each grid is determined based on the point cloud data in each grid. The grid including the static obstacle is determined as the obstacle grid, and the grid not including the static obstacle is determined as the non-obstacle grid. The static obstacle in the embodiment of the present application can be a target statically existing in the nature reserve, such as a tree.

[0060] Optionally, the point cloud data can be disturbed by sensor errors during collection, resulting in noise in the point cloud data. The collected point cloud data can be statistically filtered to eliminate the influence of noise before the point cloud data is corresponded to the grid map.

[0061] For the non-obstacle grid, the static obstacle density in the non-obstacle grid is determined based on the static obstacle in the non-obstacle grid.

[0062] Step 103, calculate the dynamic obstacle density of the non-obstacle grid based on the image data of the dynamic obstacle of the nature reserve.

[0063] The dynamic obstacle density is used to represent the distribution density of the dynamic obstacle around the grid.

[0064] In the embodiment of the present application, the dynamic obstacles in each grid are determined based on the image data in each grid, and for the obstacle-free grid, the dynamic obstacle density in the obstacle-free grid is determined based on the dynamic obstacles in the obstacle-free grid.

[0065] In a possible implementation, there may be image data of dynamic targets that affect the unmanned aerial vehicle inspection in the process of collecting the nature reserve by using a high-definition camera, for example, owls, bees, etc. The embodiment of the present application can pre-label the dynamic targets in the image data, train the collected data by using a YOLOv8 model, and obtain a target detection model.

[0066] In actual use, the target detection model can detect all dynamic targets in the detection range of the unmanned aerial vehicle and obtain the center position of all dynamic targets in the detection frame on the current image. In the process of collecting image data of dynamic obstacles of the nature reserve subsequently, a plurality of continuous frames of images can be collected by the unmanned aerial vehicle in the inspection process, the dynamic targets are tracked, and the motion trajectory of the dynamic targets is predicted, the grid of the dynamic obstacle distribution is determined based on the actual trajectory and the predicted trajectory of the dynamic targets, and then the dynamic obstacle density in the grid is calculated according to the grid where the actual trajectory and the predicted trajectory of the dynamic targets are located.

[0067] Step 104, determining a comprehensive obstacle factor of each obstacle-free grid based on the static obstacle density and the dynamic obstacle density.

[0068] In a possible implementation, the weight of the static obstacle density and the weight of the dynamic obstacle density are determined, the static obstacle density and the dynamic obstacle density are weighted and normalized based on the weights, and the comprehensive obstacle factor of each obstacle-free grid is determined.

[0069] Step 105, selecting an initial point of the grey wolf algorithm based on the static obstacle density.

[0070] In one possible implementation, the initialization of the Gray Wolf algorithm is guided by the static obstacle density in the barrier-free grid. For example, points with a static obstacle density that meets the requirements are selected as the initial points of the Gray Wolf algorithm. Optionally, the static obstacle density requirements include, but are not limited to, at least one of the following: maximum static obstacle density, or maximum sampling probability after combining static obstacle density with other influencing conditions. It should be noted that a high static obstacle density indicates that there are many targets to be monitored, such as trees, near the grid area, requiring focused monitoring of this area. Therefore, points with high static obstacle density can be prioritized to ensure focused monitoring of high-density static target areas; or points with high static obstacle density and close to the planned preset inspection path can be selected as initial points, ensuring focused monitoring of high-density static target areas while minimizing deviation from the preset inspection path.

[0071] Step 106: Perform iterative calculations using the Grey Wolf algorithm with the optimization objective of minimizing the weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path, and determine the inspection path.

[0072] In one possible implementation, after determining the initial point, a target function for the Gray Wolf Algorithm is set. This target function comprehensively considers the inspection path and the combined obstacle factors during the inspection process, minimizing the weighted sum of the inspection path and obstacle factors. This improves inspection efficiency, reduces inspection costs, and enables the drone to avoid both dynamic and static obstacles during inspection, ensuring the drone's safety. The inspection path is determined through iterative calculation based on the target function of the Gray Wolf Algorithm. After determining the inspection path, the drone performs inspections based on that path.

[0073] Based on the above technical solution, this application selects the initial point of the Gray Wolf algorithm based on the static obstacle density in each grid during the planning of UAV inspection paths. This allows for focused monitoring of high-density static target areas, thereby improving inspection efficiency and making the inspection route more reasonable. Furthermore, during the iteration of the Gray Wolf algorithm, the optimization objective is to minimize the weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path. This results in the shortest inspection route and the lowest probability of UAV collision with obstacles. Compared to existing technologies that select uniformly distributed initial points and do not consider dynamic targets when planning inspection routes, the UAV inspection optimization method for nature reserves provided in this application enables focused monitoring of high-density areas, and the planned inspection path ensures both economy and safety, thus making the planned inspection route more reasonable.

[0074] In one possible implementation, combining Figure 1 ,like Figure 2 As shown, step 102 above can be implemented through the following steps:

[0075] Step 201, determining the obstacle grid and the non-obstacle grid in the plurality of grids based on the point cloud data of the static obstacle of the nature reserve.

[0076] Step 202, clustering the obstacle grid based on a clustering algorithm to obtain a clustering area of the obstacle grid.

[0077] In a possible implementation, in the grid map, the obstacle grid is clustered by using a density-based noise application spatial clustering (DBSCAN) algorithm clustering algorithm, a plurality of clustering areas can be obtained, and in each clustering area, there are non-obstacle grids and obstacle grids. The density of the obstacle grids around the non-obstacle grid can represent the static obstacle density of the non-obstacle grid area in the current clustering area. The greater the density of the obstacle grids around the non-obstacle grid, the greater the static obstacle density of the non-obstacle grid area. The uniformity of the grey wolf algorithm initialization point can be ensured in the key inspection area through the static obstacle density. Optionally, the area marked as 1 can be clustered in the clustering process.

[0078] Step 203, calculating a first area of the obstacle grid in each clustering area, and calculating a second area of the minimum bounding rectangle of each clustering area.

[0079] Step 204, dividing the first area by the second area of each clustering area to obtain the static obstacle density of each clustering area.

[0080] As an implementation mode, the first area of each clustering area can be calculated based on opencv, the clustering area label is set for each clustering area according to the clustering area area from small to large, so as to avoid that opencv cannot distinguish different clustering areas, resulting in calculation error of the minimum bounding rectangle. The first area of the obstacle grid in each clustering area and the second area of the minimum bounding rectangle of each clustering area are calculated by using Opencv, and the clustering center coordinates of each clustering area are recorded. For each clustering area, the static obstacle density A of each clustering area is obtained by dividing the area of the obstacle grid in the clustering area by the area of the minimum bounding rectangle.

[0081] Step 205, diffusing the static obstacle density of each clustering area to the non-obstacle grid in the clustering area to obtain the static obstacle density of the non-obstacle grid.

[0082] In a possible implementation, the embodiment of the present application can diffuse the static obstacle density of each cluster region to all the unobstacle grids in the cluster region based on a two-dimensional Gaussian distribution, to obtain the static obstacle density of all the unobstacle grids in the cluster region.

[0083] Optionally, the static obstacle density of the unobstacle grid is negatively related to the distance between the unobstacle grid and the center of the cluster region, that is, the farther the unobstacle grid is from the cluster center, the smaller the static obstacle density diffused to the unobstacle region.

[0084] Based on the above technical solution, the embodiment of the present application first determines the obstacle distribution based on the point cloud data and the grid method, avoiding misjudgment of the obstacle region; secondly, the DBSCAN clustering algorithm is used to identify the cluster region of the obstacle grid; thirdly, the static obstacle density of the cluster region is calculated by the ratio of the first area to the second area, which can accurately reflect the density of the static obstacle in the region; finally, the cluster density is diffused to the unobstacle grid, so that the static obstacle density of the unobstacle grid can reflect the influence of the surrounding static obstacle, providing high-precision static obstacle data support for subsequent initial point selection and path planning, and avoiding the situation that the key area is not inspected due to inaccurate static obstacle information.

[0085] As a possible implementation, as shown in Figure 3 The step 205 can be implemented by the following steps.

[0086] Step 301: calculating the first diffusion space of the target cluster region based on the area of the target cluster region.

[0087] The target cluster region is one of the cluster regions, and the first diffusion space is positively related to the area of the target cluster region.

[0088] It can be understood that the diffusion space is a key parameter for quantifying the influence range of the static obstacle on the surrounding unobstacle grid, which is used to control the diffusion degree of the Gaussian function and represents the effective coverage range of the obstacle density decay with the spatial distance of the cluster region. The more dense the static obstacle of the cluster region (such as the tree gathering area), the larger the first area, the larger the diffusion space, and the wider the influence range of the obstacle density; the farther the unobstacle grid is from the cluster center, the stronger the attenuation effect of the diffusion space, and the lower the static obstacle density.

[0089] As an example, the first diffusion space of the cth cluster region in the plurality of cluster regions satisfies the following formula:

[0090]

[0091] wherein, represents the area of the cth cluster region.

[0092] Step 302, the first diffusion space of the target clustering area and the spatial distance between the first barrier-free grid and the target clustering area are brought into the first Gaussian distribution function, and the first Gaussian function value of the first barrier-free grid is determined.

[0093] Wherein, the first barrier-free grid is one barrier-free grid in the target clustering area.

[0094] Optionally, the first Gaussian distribution function includes a first exponential term and a first constant term, the first constant term is the reciprocal of the product of the square of the first diffusion space and 2π, and the exponent value of the first exponential term is the negative value of the quotient of the spatial distance and the square of 2 times the first diffusion space.

[0095] As an example, the first Gaussian function value of the barrier-free grid with coordinates (x, y) in the cth clustering area in the above plurality of clustering areas Satisfies the following formula:

[0096]

[0097] ( is the clustering center coordinate of the cth clustering area, and e is a natural constant.

[0098] Step 303, the product of the static obstacle density of the target clustering area and the first Gaussian function value of the first barrier-free grid is normalized to obtain the static obstacle density of the first barrier-free grid.

[0099] As an example, the normalized static obstacle density of the first barrier-free grid Satisfies the following formula:

[0100]

[0101] Wherein, Indicates the static obstacle density of the cth clustering area, Indicates min-max normalization.

[0102] It should be pointed out that the more barrier-free grids in the cth clustering area, the greater the static obstacle density of the cth clustering area, and the greater the static obstacle density of the barrier-free grid region with coordinates (x, y); the farther the distance between the barrier-free grid region with coordinates (x, y) and the clustering center of the cth clustering area, the smaller the static obstacle density of the barrier-free grid region with coordinates (x, y).

[0103] It should be noted that the above steps mainly give a technical way for the static obstacle density of the first obstacle-free grid in the target cluster area. By performing the above steps on each obstacle-free grid in each cluster area, the static obstacle density of each obstacle-free grid can be obtained.

[0104] Based on the above technical solution, the embodiment of the present application calculates the first diffusion space positively correlated with the area of the cluster area, so that the diffusion range matches the size of the cluster area, avoiding the problems of over-broad diffusion for small cluster areas or insufficient diffusion for large cluster areas. The distance between the diffusion space and the obstacle-free grid to the cluster center is substituted into the first Gaussian distribution function, and the distance attenuation characteristic of the Gaussian function is used to accurately quantify the degree of influence of the obstacle-free grid affected by the static obstacle of the cluster area (the closer the distance, the greater the influence). Finally, through normalization processing, the static obstacle density of the obstacle-free grid in different cluster areas is ensured to be comparable, avoiding the global data imbalance caused by excessively high or low density of a single cluster area. The static obstacle information that is more consistent with the actual spatial distribution is provided for the initial point selection, further ensuring the accurate identification of the key inspection area.

[0105] The above describes in detail the way of calculating the static obstacle density of each obstacle-free grid in the embodiment of the present application.

[0106] In one possible implementation manner, as shown in Figure 4 The step 103 can be implemented by the following steps.

[0107] Step 401: Collect a plurality of image frames of the nature reserve at a preset time interval, and detect dynamic targets in each image frame based on a dynamic target detection model.

[0108] As an example, the preset time interval is 2s, that is, a plurality of image frames of the nature reserve are collected at a time interval of 2s. The dynamic target detection model can be a target detection model obtained by training a single detection version 8 (You Only Look Once version 8, YOLOv8) model on the collected image frames of the nature reserve with dynamic targets.

[0109] Step 402: Determine the actual motion trajectory of the dynamic target based on a dynamic tracking algorithm, and process the actual motion trajectory based on a Kalman filtering algorithm to predict the predicted motion trajectory of the dynamic target in a future time period.

[0110] As an implementation manner, a high-definition camera is used to collect continuous frame images on a UAV inspection path, a time interval of each frame image is 2s, each collected frame image is taken as an input of a Deep Simple Online and Realtime Tracking (DeepSort) tracking algorithm, and a Kalman filter is used to predict a trajectory of a dynamic target in the image, such as predicting a motion trajectory of the dynamic target in the next 5s, usually extracting a specific position of the dynamic target in the last frame image, predicting a predicted position of the dynamic target in the next frame image, calculating a speed of the dynamic target by using an Euclidean distance / 2 between the actual position of the dynamic target in the last frame image and the specific position of the dynamic target in the current frame image, recording the predicted position coordinates and the predicted trajectory coordinates of the dynamic target in each frame image in the next 5s, and mapping the predicted position and the trajectory of the dynamic target into grid coordinates by using a Cartesian product coordinate system. Within the predicted range of the dynamic target, if the dynamic target appears, the predicted position and the predicted trajectory of the dynamic target are marked as 2 in the grid value, indicating a future dynamic obstacle region.

[0111] In step 403, the actual motion trajectory and the predicted motion trajectory are mapped into a plurality of grids, and a dynamic obstacle grid including the actual motion trajectory and / or the predicted motion trajectory and an adjacent grid with a distance less than a preset distance from the dynamic obstacle grid are determined from the plurality of grids.

[0112] As an implementation manner, the adjacent grid with a distance less than the preset distance from the dynamic obstacle grid can be a 24-neighbor grid of the dynamic obstacle grid. For example, in a square with a side length of 5 grid lengths centered on the dynamic obstacle grid, the other 24 grids except the dynamic obstacle grid are adjacent grids of the dynamic obstacle grid.

[0113] As an example, the 24-neighbor of the kth dynamic obstacle grid satisfies the following formula:

[0114]

[0115] indicates a 24-neighbor coordinate set of the kth dynamic obstacle region in the tth frame image, indicates an offset coordinate, indicates a coordinate of the kth dynamic obstacle region in the tth frame image.

[0116] In step 404, a dynamic obstacle density of a second non-obstacle grid is calculated based on a total area of the dynamic obstacle grid and the adjacent grid, and a distance between the dynamic target and the second non-obstacle grid in the adjacent grid.

[0117] As an implementation manner, the dynamic obstacle density of the second barrier-free grid in the embodiment of the present application is positively correlated with the total area of the dynamic obstacle grid and the adjacent grid, and is negatively correlated with the distance between the dynamic target and the second barrier-free grid in the adjacent grid. That is, the larger the total area is, the greater the dynamic obstacle density of the second barrier-free grid is, and the closer the distance between the second grid and the dynamic target is, the greater the dynamic obstacle density is.

[0118] In the embodiment of the present application, firstly, images are collected at preset time intervals and based on the dynamic target in the detected images, to ensure that the dynamic target (such as birds, insects) is not missed; secondly, the target tracking algorithm and the Kalman filtering algorithm are combined, so as to obtain the actual motion trajectory of the dynamic target and predict the motion trajectory in the future time period, solving the defect that the traditional algorithm only perceives the current dynamic target and cannot predict the subsequent position; then, the trajectory is mapped to the grid and the target grid and the adjacent grid are determined, to clarify the influence range of the dynamic obstacle; finally, the dynamic obstacle density is calculated based on the above information, so that the dynamic obstacle density of the barrier-free grid can reflect the real-time and future influence of the dynamic target, to provide timely and accurate dynamic obstacle data for the calculation of the comprehensive obstacle factor, to ensure that the unmanned aerial vehicle can avoid the dynamic target in advance and reduce the collision risk.

[0119] In a possible implementation manner, as shown in Figure 5 The step 404 can be implemented by the following steps.

[0120] The step 501 calculates the second diffusion space based on the total area of the dynamic obstacle grid and the adjacent grid.

[0121] The second diffusion space is positively correlated with the total area of the dynamic obstacle grid and the adjacent grid.

[0122] Optionally, the second diffusion space is calculated in a similar manner to the first diffusion space, for example, the second diffusion space of the kth dynamic obstacle grid is The following formula is met:

[0123]

[0124] The total area of the kth dynamic obstacle grid and the adjacent grid is

[0125] The step 502 brings the second diffusion space and the distance between the dynamic target and the second barrier-free grid into the second Gaussian distribution function, to determine the second Gaussian function value of the second barrier-free grid.

[0126] Optionally, the second Gaussian distribution function includes a second exponential term and a second constant term, the second constant term is the reciprocal of the product of the square of the second diffusion space and 2π, and the exponential value of the second exponential term is the negative value of the quotient of the distance between the dynamic target and the second barrier-free grid and 2 times the square of the second diffusion space.​

[0127] As an example, the second Gaussian distribution function of the non- obstacle grid in the adjacent grid of the kth dynamic obstacle region in the t-1th frame image satisfies the following formula:

[0128]

[0129] wherein (p, q) is the center coordinate of the second non- obstacle grid, and represents the coordinate difference of the dynamic target in the tth frame image and the t-1th frame image.

[0130] It should be noted that, and may be the moving distance and moving direction of the dynamic target in the tth frame image and the t-1th frame image, used to represent the moving state of the dynamic target.

[0131] Step 503, normalizing the product of the second Gaussian function value and the dynamic obstacle region indication function of the second non- obstacle grid, to obtain the dynamic obstacle density of the second non- obstacle grid.

[0132] As an example, the dynamic obstacle density of the kth second non- obstacle grid after normalization satisfies the following formula:

[0133]

[0134] is the dynamic obstacle region indication function, in the case of existing dynamic obstacle in the grid the value of is 1, in the case of non-existing dynamic obstacle in the grid the value of is 0.

[0135] In still some possible implementation manners, the dynamic obstacle density of the grid without dynamic obstacle can also be directly defaulted as 0, and the dynamic obstacle density is only calculated for the grid with dynamic obstacle, which is not limited in the present application.

[0136] Based on the above technical solution, this application embodiment calculates a second diffusion space positively correlated with the area of ​​the target grid and adjacent grids, matching the influence range of dynamic obstacles with the actual grid coverage, thus avoiding misjudgment of the influence range of dynamic obstacles. The second diffusion space and the distance from the dynamic target to the unobstructed grid are substituted into a second Gaussian distribution function, utilizing distance decay characteristics to quantify the impact of the dynamic target on unobstructed grids at different locations. A dynamic obstacle region indicator function is introduced to distinguish the presence or absence of dynamic obstacles, avoiding invalid calculations when there are no dynamic obstacles. Finally, normalization processing ensures the comparability of dynamic obstacle densities. This solution enables dynamic obstacle density to accurately reflect the real-time position and movement trend of dynamic targets on unobstructed grids, providing high-precision dynamic data for comprehensive obstacle factors, making UAV obstacle avoidance decisions more timely and accurate.

[0137] The above provides a detailed explanation of how to calculate the dynamic obstacle density of an accessible grid.

[0138] One possible implementation is, such as Figure 6 As shown, step 105 above can be implemented through the following steps:

[0139] Step 601: Set up a set of inspection route points covering nature reserves.

[0140] In one possible implementation, when setting the inspection path point set, the path point set Z={ is set based on the topography of the nature reserve, as well as the performance of the UAV's onboard camera and the inspection requirements. The set of inspection route points should include route points that can cover the entire area of ​​the nature reserve that needs to be inspected.

[0141] Step 602: Select the starting point of the inspection path from the set of inspection path points, and calculate the minimum Euclidean distance between each unobstructed grid in the adjacent grids of the grid to which the starting point of the inspection path belongs and the path point in the set of inspection path points.

[0142] Step 603: Calculate the minimum penalty distance based on the minimum Euclidean distance. The minimum penalty distance is inversely proportional to the minimum Euclidean distance.

[0143] As one possible implementation, the starting point of the inspection path is selected from the set of inspection path points. ,Will All barrier-free rasters within the 24 neighboring rasters of the given grid and Z={ The minimum Euclidean distance between all path points in} After that, the minimum penalty distance is calculated based on the minimum Euclidean distance.

[0144] As an example, minimum penalty distance Satisfy the following formula:

[0145]

[0146] wherein λ is a control dispersion factor, and the empirical value is 5.

[0147] Step 604, determining the product of the minimum penalty distance and the static obstacle density of each unobstructed grid, and selecting the unobstructed grid with the largest product as the initial point of the grey wolf algorithm.

[0148] As an implementation manner, the calculation The product of the static obstacle density and the minimum penalty distance of all unobstructed grids in the 24-neighborhood grid of the grid is calculated, and the product is normalized by using min-max to obtain the comprehensive sampling probability of each unobstructed grid. The unobstructed grid with the highest comprehensive sampling probability is selected as the initial point of the grey wolf algorithm.

[0149] Based on the above technical solutions, the embodiments of the present application set a set of inspection path points covering the entire nature reserve, avoiding the initial point deviating from the inspection range; the minimum Euclidean distance between the adjacent grid of the inspection starting point and the path point is calculated, and the minimum penalty distance inversely proportional to the distance is calculated based on the distance, so that the initial point is preferentially close to the preset path point, ensuring the continuity of the inspection path; finally, the unobstructed grid with the largest product of the minimum penalty distance and the static obstacle density is selected as the initial point, ensuring that the initial point meets the dual requirements of being close to the path and being located in the static obstacle dense area, solving the problem that the initial point of the traditional algorithm deviates from the path or misses the key area, and improving the efficiency of subsequent path planning and the inspection quality.

[0150] In a possible implementation manner, as shown in Figure 6 The step 106 can be implemented by the following steps.

[0151] Step 605, determining the objective function of the grey wolf algorithm.

[0152] The objective function is used to minimize the weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path.

[0153] As an example, the objective function H satisfies:

[0154]

[0155] wherein, is the objective function of the unmanned aerial vehicle for inspecting the nature reserve, is the inspection path length, which is obtained by accumulating the Euclidean distances between each adjacent path point; n represents the total number of path points; represents the th path point; represents the comprehensive obstacle factor of the grid where the th path point is located, and The values ​​of are all non-zero, and + =1, This is a weighting coefficient for the inspection path length. The weighting coefficients of the comprehensive obstacle factors for the inspection path. and The value can be set based on experience or other optional methods in related technologies, and this application does not limit it in this regard.

[0156] It should be pointed out that, and The settings can be tailored to different areas of a nature reserve. For example, in the core protected area of ​​a nature reserve (such as a habitat for rare birds), where static obstacles are dense and dynamic targets are active, safety is the priority in this area, and settings can be adjusted accordingly. ,like The value is 0.8. The value is set to 0.2. For buffer zones within nature reserves, where the terrain is flat and obstacles are sparse, efficiency is prioritized, and this setting is used accordingly. ,like The value is 0.2. The value is set to 0.8. For experimental areas within nature reserves, both obstacle and efficiency requirements must be met: in this case, the following settings can be adopted: and Equal or similar, such as The value is 0.5.

[0157] Step 606: Set the gray wolf population size and the maximum number of iterations, and perform iterative calculations of the gray wolf algorithm based on the objective function to select the optimal gray wolf.

[0158] During the iterative calculation process, the positions of all gray wolves are dynamically updated based on the gradient approximation. The gray wolf population size is set to 50, and the maximum number of iterations is set to 100.

[0159] In one possible implementation, if the objective function converges during the iterative calculation, the iteration can be stopped before the maximum number of iterations is reached, and the inspection path can be determined based on the current gray wolf position. If the objective function has not converged even after 100 iterations, the inspection path can be determined based on the last gray wolf position. It should be noted that the gray wolf population size and the maximum number of iterations can be set as needed, and this application does not impose any limitations on them.

[0160] Step 607: Select the grid cell to which the optimal gray wolf belongs as the grid cell in the inspection path.

[0161] Based on the technical solution, the embodiment of the present application ensures the balance between the short path and the low obstacle risk by clearly defining the minimum path length of the target function and the weighted value of the comprehensive obstacle factor; the parameters of the population size of 50 and the maximum iteration number of 100 are set, which not only avoids the insufficient optimization caused by the small population, but also avoids the low calculation efficiency caused by too many iteration numbers, and balances the optimization effect and the calculation cost; the grey wolf position is dynamically updated based on the gradient approximation value in the iteration, so that the grey wolf can adjust the position according to the spatial change of the obstacle factor, and better adapt to the distribution of dynamic and static obstacles; finally, the optimal path is composed of the grid to which the optimal grey wolf belongs, which ensures that the global optimal inspection path is obtained, and effectively improves the efficiency, safety and key area coverage of the unmanned aerial vehicle inspection, and completes the optimization target of the nature reserve inspection.

[0162] In a possible implementation, the embodiment of the present application also provides a nature reserve unmanned aerial vehicle inspection optimization system 700, as shown in Figure 7 The system comprises: an obstacle factor calculation module 701 and a path planning module 702; the obstacle factor calculation module 701 is used for dividing the nature reserve into a plurality of grids, calculating the static obstacle density of the obstacle-free grid in the plurality of grids based on the point cloud data of the static obstacles of the nature reserve, and calculating the dynamic obstacle density of the obstacle-free grid based on the image data of the dynamic obstacles of the nature reserve; wherein the static obstacle density is used to represent the distribution density of the static obstacles around the grid, and the dynamic obstacle density is used to represent the distribution density of the dynamic obstacles around the grid; the obstacle factor calculation module 701 is also used for determining the comprehensive obstacle factor of each obstacle-free grid based on the static obstacle density and the dynamic obstacle density; the path planning module 702 is used for selecting the initial point of the grey wolf algorithm based on the static obstacle density, and performing the iterative calculation of the grey wolf algorithm with the minimum weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path as the optimization target, to determine the inspection path.

[0163] In a possible implementation, the obstacle factor calculation module 701 is specifically used for: determining the obstacle grid and the obstacle-free grid in the plurality of grids based on the point cloud data of the static obstacles of the nature reserve; clustering the obstacle grid based on a clustering algorithm to obtain the clustering area of the obstacle grid; calculating the first area of the obstacle grid in each clustering area, and calculating the second area of the minimum circumscribed rectangle of each clustering area; dividing the first area of each clustering area by the second area to obtain the static obstacle density of each clustering area; and diffusing the static obstacle density of each clustering area to the obstacle-free grid in the clustering area to obtain the static obstacle density of the obstacle-free grid.

[0164] In a possible implementation, the obstacle factor calculation module 701 is specifically further configured to: calculate a first diffusion space of a target cluster region based on an area of the target cluster region, the target cluster region being one of the cluster regions, the first diffusion space being positively correlated with the area of the target cluster region; input the first diffusion space of the target cluster region and a spatial distance between the target cluster region and a first unobstacle grid into a first Gaussian distribution function to determine a first Gaussian function value of the first unobstacle grid, the first unobstacle grid being one of the unobstacle grids in the target cluster region; and normalize a product of a static obstacle density of the target cluster region and the first Gaussian function value of the first unobstacle grid to obtain the static obstacle density of the first unobstacle grid.

[0165] In a possible implementation, the obstacle factor calculation module 701 is specifically further configured to: collect a plurality of image frames of the nature reserve at preset time intervals, and detect a dynamic target in each image frame based on a dynamic target detection model; determine an actual motion trajectory of the dynamic target based on a dynamic tracking algorithm, and process the actual motion trajectory based on a Kalman filtering algorithm to predict a predicted motion trajectory of the dynamic target in a future time period; map the actual motion trajectory and the predicted motion trajectory into the plurality of grids, and determine, from the plurality of grids, a dynamic obstacle grid including the actual motion trajectory and / or the predicted motion trajectory, and an adjacent grid having a distance less than a preset distance from the dynamic obstacle grid; and calculate a dynamic obstacle density of a second unobstacle grid based on a total area of the dynamic obstacle grid and the adjacent grid, and a distance between the dynamic target and a second unobstacle grid in the adjacent grid.

[0166] In a possible implementation, the obstacle factor calculation module 701 is specifically further configured to: calculate a second diffusion space based on a total area of the dynamic obstacle grid and the adjacent grid, the second diffusion space being positively correlated with the area of the dynamic obstacle grid and the adjacent grid; input the second diffusion space and the distance between the dynamic target and the second unobstacle grid into a second Gaussian distribution function to determine a second Gaussian function value of the second unobstacle grid; and normalize a product of the second Gaussian function value and a dynamic obstacle region indicator function of the second unobstacle grid to obtain the dynamic obstacle density of the second unobstacle grid.

[0167] In a possible implementation, the path planning module 702 is further configured to: set a set of inspection path points covering the nature reserve; select an inspection path starting point from the set of inspection path points, and calculate the minimum Euclidean distance between each unobstructed grid in the adjacent grid of the grid to which the inspection path starting point belongs and the path points in the set of inspection path points; based on the minimum Euclidean distance, calculate a minimum penalty distance, which is inversely proportional to the minimum Euclidean distance; determine the product of the minimum penalty distance of each unobstructed grid and the static obstacle density, and select the unobstructed grid with the maximum product as the initial point of the grey wolf algorithm.

[0168] In a possible implementation, the path planning module 702 is further configured to: determine a target function of the grey wolf algorithm, the target function being used to minimize the length of the inspection path and the weighted value of the comprehensive obstacle factor of the inspection path; set the size of the grey wolf population and the maximum number of iterations, and perform iterative calculation of the grey wolf algorithm based on the target function to select an optimal grey wolf; in the process of iterative calculation, the positions of all the grey wolves are dynamically updated based on gradient approximation, the size of the grey wolf population is set to 50, and the maximum number of iterations is set to 100; and the grid to which the optimal grey wolf belongs is taken as a grid in the inspection path.

[0169] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0170] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment mainly describes the difference from other embodiments.

Claims

1. A natural reserve unmanned aerial vehicle inspection optimization method, characterized in that, The method comprises: dividing the nature reserve into a plurality of grids, calculating a static obstacle density of an unobstructed grid in the plurality of grids based on point cloud data of static obstacles of the nature reserve, and calculating a dynamic obstacle density of the unobstructed grid based on image data of dynamic obstacles of the nature reserve; wherein the static obstacle density is used to represent the distribution density of static obstacles around the grid, and the dynamic obstacle density is used to represent the distribution density of dynamic obstacles around the grid; determining a comprehensive obstacle factor of each unobstructed grid based on the static obstacle density and the dynamic obstacle density; selecting an initial point of the grey wolf algorithm based on the static obstacle density, and performing iterative calculation of the grey wolf algorithm with the minimum weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path as the optimization objective to determine the inspection path; wherein the initial point of the grey wolf algorithm is a point with the maximum static obstacle density or a point with the maximum product of the static obstacle density and the minimum penalty distance; the minimum penalty distance is inversely proportional to the minimum Euclidean distance from the point to the inspection path; wherein the calculation of the static obstacle density of the unobstructed grid in the plurality of grids based on the point cloud data of the static obstacles of the nature reserve comprises: determining the obstructed grid and the unobstructed grid in the plurality of grids based on the point cloud data of the static obstacles of the nature reserve; clustering the obstructed grids based on a clustering algorithm to obtain a clustering area of the obstructed grids; calculating a first area of the obstructed grids in each clustering area, and calculating a second area of a minimum circumscribed rectangle of each clustering area; dividing the first area of each clustering area by the second area to obtain a static obstacle density of each clustering area; diffusing the static obstacle density of each clustering area to the unobstructed grids in the clustering area to obtain the static obstacle density of the unobstructed grids. 2.The method of claim 1, wherein, The diffusing the static obstacle density of each clustering area to the unobstructed grids in the clustering area to obtain the static obstacle density of the unobstructed grids comprises: calculating a first diffusion space of a target clustering area based on the area of the target clustering area; wherein the target clustering area is one of the clustering areas, and the first diffusion space is positively correlated with the area of the target clustering area; inputting the first diffusion space of the target clustering area and a spatial distance between a first unobstructed grid and the target clustering area into a first Gaussian distribution function to determine a first Gaussian function value of the first unobstructed grid; the first unobstructed grid is one of the unobstructed grids in the target clustering area; normalizing the product of the static obstacle density of the target clustering area and the first Gaussian function value of the first unobstructed grid to obtain the static obstacle density of the first unobstructed grid. 3.The method of claim 2, wherein, The first Gaussian distribution function comprises a first exponential term and a first constant term, the first constant term is the reciprocal of the product of the square of the first diffusion space and 2π, and the exponent value of the first exponential term is the negative value of the quotient of the spatial distance and 2 times the square of the first diffusion space. 4.The method of claim 1, wherein, The image data of the dynamic obstacle based on the nature reserve is used to calculate the dynamic obstacle density of the barrier-free grid, comprising: a plurality of image frames of the nature reserve are collected at preset time intervals, and a dynamic target in each image frame is detected based on a dynamic target detection model; an actual motion trajectory of the dynamic target is determined based on a dynamic tracking algorithm, and the actual motion trajectory is processed based on a Kalman filtering algorithm to predict a predicted motion trajectory of the dynamic target in a future time period; the actual motion trajectory and the predicted motion trajectory are mapped into the plurality of grids, and a dynamic obstacle grid including the actual motion trajectory and / or the predicted motion trajectory is determined from the plurality of grids, and an adjacent grid with a distance less than a preset distance from the dynamic obstacle grid is determined; a total area of the dynamic obstacle grid and the adjacent grid is used to calculate a dynamic obstacle density of a second barrier-free grid in the adjacent grid based on a distance between the dynamic target and the second barrier-free grid. 5.The method of claim 4, wherein, The area of the dynamic obstacle grid and the adjacent grid is used to calculate the dynamic obstacle density of the second barrier-free grid based on the distance between the dynamic target and the second barrier-free grid in the adjacent grid, comprising: a second diffusion space is calculated based on the total area of the dynamic obstacle grid and the adjacent grid, and the second diffusion space is positively correlated with the area of the dynamic obstacle grid and the adjacent grid; the second diffusion space and the distance between the dynamic target and the second barrier-free grid are brought into a second Gaussian distribution function to determine a second Gaussian function value of the second barrier-free grid; the product of the second Gaussian function value and a dynamic obstacle area indication function of the second barrier-free grid is normalized to obtain the dynamic obstacle density of the second barrier-free grid. 6.The method of claim 5, wherein, The second Gaussian distribution function comprises a second exponential term and a second constant term, the second constant term is the reciprocal of the product of the square of the second diffusion space and 2π, and the exponential value of the second exponential term is the negative value of the quotient of the distance between the dynamic target and the second barrier-free grid and twice the square of the second diffusion space. 7.The method of claim 1, wherein, The initial point of the grey wolf algorithm is selected based on the static obstacle density, comprising: a set of inspection path points covering the nature reserve is set; an inspection path starting point is selected from the set of inspection path points, and the minimum Euclidean distance between each barrier-free grid in the adjacent grid of the grid to which the inspection path starting point belongs and the path points in the set of inspection path points is calculated; based on the minimum Euclidean distance, a minimum penalty distance is calculated, and the minimum penalty distance is inversely proportional to the minimum Euclidean distance; the product of the minimum penalty distance of each barrier-free grid and the static obstacle density is determined, and the barrier-free grid with the maximum product is selected as the initial point of the grey wolf algorithm. 8.The method of claim 1, wherein, The iteration calculation of the grey wolf algorithm is performed with the minimum weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path as the optimization target to determine the inspection path, comprising: a target function of the grey wolf algorithm is determined, and the target function is used to minimize the weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path; Set the gray wolf population size and the maximum number of iterations, and perform iterative calculation of the gray wolf algorithm based on the target function to select the optimal gray wolf; during the iterative calculation, the positions of all gray wolves are dynamically updated based on gradient approximation values, the gray wolf population size is set to 50, and the maximum number of iterations is set to 100; The grid to which the optimal gray wolf belongs is taken as a grid in the inspection path.

9. A natural reserve unmanned aerial vehicle inspection optimization system, characterized in that, The system comprises an obstacle factor calculation module and a path planning module; The obstacle factor calculation module is configured to divide the nature reserve into a plurality of grids, calculate a static obstacle density of an unobstructed grid in the plurality of grids based on point cloud data of static obstacles of the nature reserve, and calculate a dynamic obstacle density of the unobstructed grid based on image data of dynamic obstacles of the nature reserve; wherein the static obstacle density is used to represent a distribution density of static obstacles around the grid, and the dynamic obstacle density is used to represent a distribution density of dynamic obstacles around the grid; The obstacle factor calculation module is further configured to determine a comprehensive obstacle factor of each unobstructed grid based on the static obstacle density and the dynamic obstacle density. The path planning module is configured to select an initial point of the gray wolf algorithm based on the static obstacle density, and perform iterative calculation of the gray wolf algorithm with a minimum weighted value of the inspection path length and the comprehensive obstacle factor of the inspection path as an optimization target to determine the inspection path; wherein the initial point of the gray wolf algorithm is a point with the maximum static obstacle density or a point with the maximum product of the static obstacle density and a minimum penalty distance; the minimum penalty distance is inversely proportional to a minimum Euclidean distance of a point to the inspection path. The obstacle factor calculation module is further configured to determine unobstructed grids and obstructed grids in the plurality of grids based on the point cloud data of the static obstacles of the nature reserve, cluster the obstructed grids based on a clustering algorithm to obtain a clustering area of the obstructed grids, calculate a first area of the obstructed grids in each clustering area, and calculate a second area of a minimum circumscribed rectangle of each clustering area; divide the first area of each clustering area by the second area to obtain a static obstacle density of each clustering area; and diffuse the static obstacle density of each clustering area to the unobstructed grids in the clustering area to obtain the static obstacle density of the unobstructed grids.

Images