Low-altitude forestry inspection target visual positioning method and system
By detecting nonlinear distortion during low-altitude forestry inspections and adaptively setting the grid subdivision scale, a spherical mapping relationship is constructed. Combined with a deep learning model, the problem of inaccurate target positioning in complex terrain using traditional methods is solved, and high-precision target positioning is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-03
- Publication Date
- 2026-03-13
AI Technical Summary
Traditional low-altitude forestry inspection methods suffer from nonlinear distortion in complex terrain, leading to inaccurate target positioning and failing to meet the needs of modern forestry's refined management and emergency response.
By acquiring POS and pose data from inspection images, detecting the degree of nonlinear distortion, adaptively setting the mesh subdivision scale, constructing a spherical mesh mapping relationship, and combining a deep learning model to identify target locations, accurate positioning of pixels and ground coordinates is achieved.
It improves the accuracy and reliability of the spatial location of low-altitude forestry inspection targets, reduces the nonlinear distortion deviation caused by traditional planar projection, and improves the accuracy of target positioning.
Smart Images
Figure 2644E744-80C4-4749-8DD2-8FC37524F058 
Figure 5391AA3D-455F-4C6F-B925-5B75C12CBD5F 
Figure 9F79E183-495A-467A-A95A-7505E838DDF4
Abstract
Description
Technical Field
[0001] This invention relates to the field of geometric image transformation technology, which projects images onto non-planar surfaces, and specifically to a visual positioning method and system for low-altitude forestry inspection targets. Background Technology
[0002] With the increasing efforts in national forestry resource protection and the advancement of smart forestry construction, low-altitude drone inspections, characterized by their high mobility and wide coverage, have gradually become an important technical means for forestry fire monitoring, pest and disease investigation, and resource inventory. However, drone inspections generally lack accurate positioning of forestry inspection targets. Traditional vision-based low-altitude target positioning methods mainly rely on direct georeferenced reference technology, which involves acquiring high-precision position and attitude data of the aircraft through satellite navigation systems and inertial measurement units, combining this data with camera parameters to construct an imaging geometric model, and directly projecting the inspection image coordinates onto the ground. This method employs the planar Earth assumption or uses a digital elevation model for correction.
[0003] Traditional low-altitude target positioning methods have significant limitations in complex terrains such as mountains and hills. Because forestry patrols often employ low-altitude flights and use large-angle, wide-angle lenses, traditional planar projection is prone to nonlinear distortion, leading to deviations when calculating target points based on the line of sight. Therefore, there is an urgent need for a low-altitude target positioning method that can adapt to large-angle imaging and complex terrain, improving the accuracy and reliability of the spatial geographic location of targets in low-altitude forestry patrols, and meeting the pressing needs of modern forestry's refined management and emergency response. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention proposes a visual positioning method and system for low-altitude forestry inspection targets, which can improve the accuracy and reliability of the spatial geographic location of low-altitude forestry inspection targets. The specific technical solution is as follows: Firstly, a visual positioning method for low-altitude forestry inspection targets is provided. In a first feasible implementation of this method, the method includes: Acquire inspection images of the inspection area, as well as POS data and attitude data of the camera during image acquisition; The degree of nonlinear distortion of the inspection image projected onto the ground is detected, and the mesh subdivision scale is set according to the degree of nonlinear distortion. The field of view sphere constructed based on the camera optical center is divided according to the mesh subdivision scale, and the spherical mesh corresponding to the inspection image is determined according to the shooting parameters, POS data and attitude data. By combining the camera's shooting parameters, POS data, and attitude data, the pixel-spherical mapping relationship between pixels and spherical grids in the inspection image, as well as the ground-spherical mapping relationship between ground coordinates and spherical grids within the inspection area, are determined. Identify and locate the pixel position of the forestry inspection target in the inspection image, and determine the ground coordinates corresponding to the forestry inspection target by combining the pixel-spherical mapping relationship and the ground-spherical mapping relationship.
[0005] In conjunction with the first possible implementation of the first aspect, in the second possible implementation of the first aspect, detecting the degree of nonlinear distortion of the inspection image projected onto the ground includes: The center pixel position of the forestry inspection target in the inspection image is located using a target detection algorithm; Based on the shooting parameters, locate the pixel position corresponding to the optical center of the camera, and combine it with the pixel position of the center point of the forestry inspection target to determine the degree of nonlinear distortion of the inspection image.
[0006] In conjunction with the first feasible method of the first aspect, in the third feasible method of the first aspect, the mesh subdivision scale is set according to the degree of nonlinear distortion, including: The horizontal and vertical field of view of the camera are calculated based on the shooting parameters, and the horizontal and vertical angle steps are set according to the degree of nonlinear distortion. By combining the horizontal field of view, vertical field of view, horizontal angle step size, and vertical angle step size, the number of horizontal grids and the number of vertical grids are calculated, and the grid subdivision scale is determined.
[0007] In conjunction with the first feasible method of the first aspect, in the fourth feasible method of the first aspect, the spherical grid corresponding to the inspection image is determined based on the shooting parameters, POS data, and attitude data, including: The horizontal and vertical field of view of the camera when capturing the inspection image are calculated based on the shooting parameters, and the field of view of the camera during shooting is determined by the horizontal and vertical field of view. Using the frustum range, POS data, and attitude data, the spherical grid corresponding to the inspection image in the field of view grid is determined.
[0008] In conjunction with the first implementable method of the first aspect, the fifth implementable method of the first aspect determines the pixel-spherical mapping relationship between pixels in the inspection image and the spherical grid, including: Based on the camera's shooting parameters, POS data, and attitude data, determine the pixel position of the camera's optical center in the inspection image, as well as the corresponding spherical grid in the field of view sphere. Based on the pixel position of the camera's optical center, the inspection image is divided into grids according to the grid subdivision scale, and the pixel grid corresponding to the camera's optical center is determined. Based on the pixel grid and spherical grid corresponding to the optical center of the camera, the mapping relationship between the pixel grid and the spherical grid is determined, and the pixel-spherical mapping relationship is obtained.
[0009] In conjunction with the first feasible method of the first aspect, the sixth feasible method of the first aspect determines the ground-spherical mapping relationship between the ground coordinates of the inspection area and the spherical grid, including: Based on the camera's shooting parameters, POS data, and attitude data, determine the ground point corresponding to the camera's optical center in the inspection area, as well as the corresponding spherical grid in the field of view grid sphere. Based on the ground point coordinates corresponding to the camera optical center, the ground in the inspection area is divided into grids according to the grid subdivision scale, and the ground grid corresponding to the camera optical center is determined. Based on the ground grid and spherical grid corresponding to the camera's optical center, the mapping relationship between the ground grid and the spherical grid is determined, thus obtaining the ground-spherical mapping relationship.
[0010] In conjunction with the first possible implementation of the first aspect, in the seventh possible implementation of the first aspect, identifying and locating the pixel position of the forestry inspection target in the inspection image includes: The trained deep learning model identifies forestry inspection targets in the inspection images and locates the pixel positions of the forestry inspection targets.
[0011] In conjunction with the first feasible method of the first aspect, the eighth feasible method of the first aspect determines the ground coordinates corresponding to the forestry inspection target, including: Based on the pixel position of the forestry inspection target, the spherical grid corresponding to the forestry inspection target is determined through the pixel-spherical mapping relationship; Based on the spherical grid corresponding to the forestry inspection target, the ground grid corresponding to the forestry inspection target is determined through the ground-spherical mapping relationship, and the ground coordinates of the forestry inspection target are located.
[0012] In conjunction with the first implementable method of the first aspect, the ninth implementable method of the first aspect also includes: Obtain a real-world 3D model of the inspection area, and determine the actual elevation of the forestry inspection target by combining the ground coordinates of the target.
[0013] Secondly, a low-altitude forestry inspection target visual positioning system is provided, including: The data acquisition module is configured to acquire inspection images of the inspection area, POS data and attitude data of the camera during image acquisition; The distortion detection module is configured to detect the degree of nonlinear distortion after the inspection image is projected onto the ground, and to set the mesh subdivision scale according to the degree of nonlinear distortion. The spherical meshing module is configured to mesh the field of view sphere constructed based on the camera optical center according to the meshing scale, and determine the spherical mesh corresponding to the inspection image according to the shooting parameters, POS data and attitude data. The mapping module is configured to combine the camera's shooting parameters, POS data, and attitude data to determine the pixel-spherical mapping relationship between pixels and spherical grids in the inspection image, as well as the ground-spherical mapping relationship between ground coordinates and spherical grids within the inspection area. The target localization module is configured to identify and locate the pixel position of the forestry inspection target in the inspection image, and determine the ground coordinates corresponding to the forestry inspection target by combining the pixel-spherical mapping relationship and the ground-spherical mapping relationship.
[0014] Beneficial Effects: The low-altitude forestry inspection target visual positioning method and system of this invention can adaptively adjust the grid subdivision scale based on the degree of nonlinear distortion caused by projecting inspection images onto the ground using traditional planar projection methods. Based on the adjusted grid subdivision scale, the spherical field of view when the camera captures the inspection image, as well as the inspection image and the inspection area, are rationally subdivided, thereby distributing the nonlinear distortion evenly across each grid and reducing projection-induced deviations. Using the subdivided spherical grid as an intermediate reference, the pixel-to-spherical mapping relationship between pixels in the inspection image and the spherical grid is used to map the identified forestry inspection target pixel positions in the inspection image to the corresponding spherical grid. Furthermore, the ground coordinates of the forestry inspection target are located through the ground-to-spherical mapping relationship between ground coordinates and the spherical grid. This solves the problem of nonlinear distortion easily caused by traditional planar projection, improving the accuracy and reliability of the spatial geographical location of low-altitude forestry inspection targets. Attached Figure Description
[0015] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.
[0016] Figure 1 A flowchart of a low-altitude forestry inspection target visual positioning method provided in an embodiment of the present invention; Figure 2 This is a system block diagram of a low-altitude forestry inspection target visual positioning system provided in an embodiment of the present invention; Figure 3 The field-of-view grid spherical surface is obtained by spherical subdivision; Figure 4 This is a schematic diagram illustrating the mapping relationship between spherical meshes and pixel meshes; Figure 5 This is a schematic diagram illustrating the mapping relationship between spherical meshes and ground meshes. Detailed Implementation
[0017] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.
[0018] It should be understood that when inspecting forested areas, drones equipped with cameras featuring large tilt angles and wide-angle lenses can fly at low altitudes along the inspection path, capturing images in real time. Simultaneously, the camera's shooting parameters, POS data, and attitude data are recorded. Each image captured by the camera corresponds to a different inspection area. Shooting parameters include focal length, image size, and resolution; POS data includes the camera's spatial geographic coordinates, flight direction, and flight speed; and attitude data includes heading angle, pitch angle, and roll angle.
[0019] Example 1 like Figure 1 The flowchart shown illustrates a visual positioning method for low-altitude forestry inspection targets. This positioning method includes: Step 1: Acquire inspection images of the inspection area, as well as the camera's POS data and attitude data during image acquisition; Step 2: Detect the degree of nonlinear distortion of the inspection image projected onto the ground, and set the mesh subdivision scale according to the degree of nonlinear distortion; Step 3: Divide the field of view sphere constructed based on the camera optical center according to the mesh subdivision scale, and determine the spherical mesh corresponding to the inspection image according to the shooting parameters, POS data and attitude data; Step 4: Combining the camera's shooting parameters, POS data, and attitude data, determine the pixel-spherical mapping relationship between pixels and spherical grids in the inspection image, and the ground-spherical mapping relationship between ground coordinates and spherical grids within the inspection area; Step 5: Identify and locate the pixel position of the forestry inspection target in the inspection image, and determine the ground coordinates corresponding to the forestry inspection target by combining the pixel-spherical mapping relationship and the ground-spherical mapping relationship.
[0020] Specifically, firstly, the system acquires inspection images of the inspection area captured by the camera, along with POS and attitude data from when the camera captured these images. Then, it detects the degree of nonlinear distortion in the projection of the inspection images onto the Earth's plane and adaptively sets the meshing scale for subsequent mesh generation based on this distortion. Next, a spherical field of view is constructed with the camera's optical center as its center, and the spherical field of view is meshed based on the set meshing scale. Finally, the corresponding spherical mesh for the inspection images is determined based on the camera's shooting parameters, POS data, and attitude data, thereby distributing the nonlinear distortion caused by projection evenly across all meshes and reducing projection-induced deviations.
[0021] Then, based on the camera's shooting parameters, POS data, and attitude data during image capture, the pixel-to-spherical mapping relationship between the pixels of the inspection image and the spherical grid of the field of view, as well as the ground-to-spherical mapping relationship between the ground coordinates of the inspection area and the spherical grid, can be determined. Finally, the pixel positions of forestry inspection targets in the inspection image can be identified and located. Using the spherical grid as an intermediate reference, the pixel positions of the forestry inspection targets are mapped onto the segmented field of view sphere through the pixel-to-spherical mapping relationship. Then, the ground coordinates corresponding to the forestry inspection targets are located through the ground-to-spherical mapping relationship. This solves the problem of nonlinear distortion easily caused by traditional planar projection and improves the accuracy and reliability of the spatial geographical location of low-altitude forestry inspection targets.
[0022] In this embodiment, optionally, detecting the degree of nonlinear distortion of the inspection image projected onto the ground includes: The center pixel position of the forestry inspection target in the inspection image is located using a target detection algorithm; Based on the shooting parameters, locate the pixel position corresponding to the optical center of the camera, and combine it with the pixel position of the center point of the forestry inspection target to determine the degree of nonlinear distortion of the inspection image.
[0023] Specifically, firstly, existing target detection algorithms, such as the YOLO algorithm, can be used to detect forestry inspection targets in the inspection image and accurately locate the center pixel position of the forestry inspection target. The YOLO algorithm is a conventional technique in this field, and its specific steps will not be described in detail here.
[0024] Then, based on the shooting parameters when the camera captures the inspection image, the pixel position corresponding to the projection of the camera's optical center along the optical axis into the inspection image can be located, i.e., the principal point of the image. The specific calculation formula is as follows: in, The pixel position corresponding to the camera's optical center. , These represent the total number of horizontal pixels and the total number of vertical pixels in the shooting parameters, respectively.
[0025] After calculating the pixel position of the camera's optical center, since the pixels corresponding to the camera's optical center in the inspection image do not undergo deformation, the degree of nonlinear distortion of the inspection image projected onto the ground can be calculated based on the pixel position of the camera's optical center. The degree of nonlinear distortion includes the degree of pixel distortion in the horizontal direction, the degree of pixel distortion in the vertical direction, the degree of attitude distortion in the horizontal direction, and the degree of attitude distortion in the vertical direction. The specific calculation formula is as follows: ; ; ; .
[0026] in, The physical length in the horizontal direction of a pixel. The physical length in the vertical direction of a pixel. , These represent the degree of pixel distortion in the horizontal direction and the degree of pixel distortion in the vertical direction, respectively. , These represent the degree of attitude distortion in the horizontal direction and the degree of attitude distortion in the vertical direction, respectively. The center pixel location of the forestry inspection target. Focal length For heading angle, It is the pitch angle.
[0027] In this embodiment, optionally, the mesh subdivision scale is set according to the degree of nonlinear distortion, including: The horizontal and vertical field of view of the camera are calculated based on the shooting parameters, and the horizontal and vertical angle steps are set according to the degree of nonlinear distortion. By combining the horizontal field of view, vertical field of view, horizontal angle step size, and vertical angle step size, the number of horizontal grids and the number of vertical grids are calculated, and the grid subdivision scale is determined.
[0028] Specifically, firstly, the horizontal and vertical field of view of the camera can be calculated based on the focal length, horizontal frame, and vertical frame when the camera captures the inspection images. The specific calculation formula is as follows: ; ; in, , These are the horizontal field of view and the vertical field of view, respectively. , These are horizontal and vertical aspect ratios, respectively.
[0029] Simultaneously, the corresponding horizontal and vertical angle step sizes can be reasonably set based on the detected degree of nonlinear distortion. The specific calculation formula is as follows: ; ; in, , These are the horizontal angular step size and the vertical angular step size, respectively. This represents the number of forestry inspection targets in the inspection images.
[0030] Then, by combining the horizontal field of view, vertical field of view, horizontal angle step size, and vertical angle step size, the number of horizontal and vertical grids can be calculated as the grid subdivision scale. The specific calculation formula is as follows: ; ; in, , These represent the number of horizontal grid cells and the number of vertical grid cells, respectively.
[0031] In this embodiment, optionally, determining the spherical grid corresponding to the inspection image based on the shooting parameters, POS data, and attitude data includes: The horizontal and vertical field of view of the camera when capturing the inspection image are calculated based on the shooting parameters, and the field of view of the camera during shooting is determined by the horizontal and vertical field of view. Using the frustum range, POS data, and attitude data, the spherical grid corresponding to the inspection image in the field of view grid is determined.
[0032] Specifically, after determining a reasonable mesh partitioning scale, the optical center of the camera during image capture can be used as the sphere center, and a cubic spherical projection method can be used to construct the field of view sphere. Then, based on the set mesh partitioning scale, a ray mesh method is used to partition the field of view sphere, generating a mesh as shown below. Figure 3 The field-of-view grid sphere is shown. Since the resulting field-of-view grid sphere includes some spherical grids corresponding to non-inspection images, the horizontal and vertical field-of-view angles of the camera when the inspection image was captured can be calculated using the above calculation method. This allows us to determine the frustum range corresponding to the inspection image, specifically the azimuth angle. Zenith .
[0033] Then take the camera's optical center as Based on the camera's POS and attitude data, the camera's real spatial position is located. Using the real spatial position as the center of a sphere, and combining the calculated azimuth and zenith angle ranges, the spherical grid within the aforementioned field of view range is obtained, which is the spherical grid corresponding to the inspection image.
[0034] In this embodiment, optionally, determining the pixel-to-sphere mapping relationship between pixels in the inspection image and the spherical grid includes: Based on the camera's shooting parameters, POS data, and attitude data, determine the pixel position of the camera's optical center in the inspection image, as well as the corresponding spherical grid in the field of view sphere. Based on the pixel position of the camera's optical center, the inspection image is divided into grids according to the grid subdivision scale, and the pixel grid corresponding to the camera's optical center is determined. Based on the pixel grid and spherical grid corresponding to the optical center of the camera, the mapping relationship between the pixel grid and the spherical grid is determined, and the pixel-spherical mapping relationship is obtained.
[0035] Specifically, such as Figure 4 As shown, after determining the spherical grid corresponding to the inspection image, the spherical grid where the camera optical center is located on the field of view can be located along the center line from the camera optical center, and the pixel position of the camera optical center in the inspection image can be located according to the shooting parameters when the inspection image is taken.
[0036] Because traditional methods of projecting inspection images onto the ground plane do not deform the pixels corresponding to the camera's optical center in the inspection image, the camera's optical center can be used as a mapping reference point to determine the pixel-spherical mapping relationship. Specifically, firstly, using the pixel position where the camera's optical center is located in the inspection image as the center, and combining the image partitioning scale calculated based on the shooting parameters and the previously set grid partitioning scale, the inspection image is divided into grids to obtain the grid image corresponding to the inspection area. The specific calculation formula for the image partitioning scale is as follows: , ; in, , These are the horizontal and vertical division intervals, respectively.
[0037] Then, based on the relative positional relationship between other pixel grids in the grid image and the pixel grid to which the camera optical center belongs, as well as the relative positional relationship between the spherical grid to which the camera optical center belongs and other spherical grids, the pixel-spherical mapping relationship between each pixel grid and the spherical grid in the inspection image can be determined.
[0038] In this embodiment, optionally, determining the ground-spherical mapping relationship between the ground coordinates of the inspection area and the spherical grid includes: Based on the camera's shooting parameters, POS data, and attitude data, determine the ground point corresponding to the camera's optical center in the inspection area, as well as the corresponding spherical grid in the field of view grid sphere. Based on the ground point coordinates corresponding to the camera optical center, the ground in the inspection area is divided into grids according to the grid subdivision scale, and the ground grid corresponding to the camera optical center is determined. Based on the ground grid and spherical grid corresponding to the camera's optical center, the mapping relationship between the ground grid and the spherical grid is determined, thus obtaining the ground-spherical mapping relationship.
[0039] Specifically, since the pixels corresponding to the camera's optical center in the inspection image do not deform when projected onto the ground plane using traditional methods, the camera's optical center is used as a mapping reference point to determine the ground-sphere mapping relationship.
[0040] Specifically, such as Figure 5 As shown, firstly, the camera's geographical coordinates are located based on the camera's shooting parameters, POS data, and attitude data during the inspection image capture. Then, based on the camera's geographical coordinates, the intersection point between the camera's optical center and the ground of the real-world 3D model of the forest-covered area is searched along the center line of the camera's field of view. The determined intersection point is the ground point corresponding to the camera's optical center. The coordinates of the ground point are: ; ; ; ; in, The geographic coordinates of the ground point The geographic coordinates of the camera's optical center. Let be the direction vector of the optical axis in the geodetic coordinate system. The distance between the optical center of the camera and the ground point. The height of the camera's optical center above the Earth's surface. Let be the radius of curvature of the ellipsoid's prime meridian. Let be the radius of curvature of the meridian. For the camera pose rotation matrix, For roll angle, Based on the calculated latitude and longitude coordinates of the ground points The elevation values of the ground points retrieved.
[0041] Then, the line of sight is extended based on the camera's view frustum range to determine the ground area corresponding to the inspection image, i.e., the inspection area corresponding to the inspection image. Next, the inspection area can be divided into grids using the ground point corresponding to the camera's optical center as the center, according to the ground division scale calculated based on the grid partitioning scale. The ground grid containing the ground point in the divided ground grid is the ground grid corresponding to the camera's optical center. The ground partitioning scale includes the horizontal grid step size and the vertical grid step size, and their specific calculation formulas are as follows: , ; in, , These are the horizontal grid step size and the vertical grid step size, respectively. , These represent the maximum horizontal distance and the maximum vertical distance, respectively, and the specific calculation formulas are as follows: ; .
[0042] After the inspection area is divided into grids, the mapping relationship between each ground grid and the spherical grid in the inspection area can be determined by the relative positional relationship between the ground grid to which the ground point corresponding to the camera optical center belongs and other ground grids, as well as the relative positional relationship between the spherical grid to which the camera optical center belongs and other spherical grids.
[0043] In this embodiment, optionally, identifying and locating the pixel position of the forestry inspection target in the inspection image includes: The trained deep learning model identifies forestry inspection targets in the inspection images and locates the pixel positions of the forestry inspection targets.
[0044] Specifically, a pre-trained deep learning model can quickly and accurately identify forestry inspection targets in inspection images and automatically label them, improving the image positioning accuracy of forestry inspection targets, thereby improving the accuracy of the aforementioned grid subdivision scale and the ground coordinate positioning accuracy of forestry inspection targets. In this embodiment, YOLO and Transformer can be used as deep learning models for locating the pixel positions of forestry inspection targets.
[0045] In this embodiment, optionally, determining the ground coordinates corresponding to the forestry inspection target includes: Based on the pixel position of the forestry inspection target, the spherical grid corresponding to the forestry inspection target is determined through the pixel-spherical mapping relationship; Based on the spherical grid corresponding to the forestry inspection target, the ground grid corresponding to the forestry inspection target is determined through the ground-spherical mapping relationship, and the ground coordinates of the forestry inspection target are located.
[0046] Specifically, after using a deep learning model to identify and locate the pixel positions of forestry inspection targets in the inspection images, the corresponding pixel grid can be determined based on the pixel positions. Then, based on the pixel-to-spherical mapping relationship between the pixel grid and the spherical grid, the corresponding spherical grid can be determined. Next, based on the ground-to-spherical mapping relationship between the spherical grid and the ground grid, the corresponding ground grid can be determined. Finally, based on the ground coordinates of the corresponding ground grid, the ground coordinates of the forestry target can be determined.
[0047] Specifically, firstly, based on the pixel position of the forestry inspection target in the inspection image and the pixel position of the camera's optical center, the pixel offset of the forestry inspection target relative to the camera's optical center is calculated. The specific calculation formula is as follows: , ; in, , These are the horizontal pixel offset and the vertical pixel offset, respectively. , These represent the pixel positions corresponding to the forestry inspection target and the camera's optical center, respectively.
[0048] Then, based on the horizontal and vertical pixel offsets, and combined with the previously set image segmentation scale, the horizontal and vertical grid indices corresponding to the forestry inspection targets can be calculated. The specific calculation formulas are as follows: , ; Then, based on the horizontal and vertical grid indices, combined with the set horizontal and vertical angle steps, and the pixel offset of the forestry inspection target relative to the camera's optical center, the azimuth angle of the spherical grid corresponding to the forestry inspection target is calculated. and zenith The specific calculation formula is as follows: , ; in, , These are the horizontal and vertical starting azimuth angles, respectively, and the specific calculation formulas are as follows: , ; , These are the horizontal and vertical proportions, and the specific calculation formulas are as follows: , .
[0049] Finally, based on the calculated horizontal and vertical proportions, combined with the set ground division scale, the offset of the forestry inspection target relative to the camera's optical center on the ground point within the inspection area can be calculated. The specific calculation formula is as follows: , ; Then, based on the azimuth angle of the spherical grid corresponding to the forestry inspection target... and zenith By combining the offset of the inspection target and the ground coordinates of the ground points, as well as the attitude data of the camera when capturing the inspection images, the ground coordinates of the inspection target are calculated. The specific calculation formula is as follows: ; .
[0050] Example 2 Example 2 is largely the same as Example 1, except that in this example, it further includes: Obtain a real-world 3D model of the inspection area, and determine the actual elevation of the forestry inspection target by combining the ground coordinates of the target.
[0051] Specifically, after calculating the ground coordinates of the inspection target point, the digital elevation model or real-scene 3D model corresponding to the inspection area can be queried based on the ground coordinates. The elevation value of the location of the inspection target point can be determined through the digital elevation model or real-scene 3D model, and combined with the ground coordinates of the inspection target point to form the real geographic coordinates of the inspection target point.
[0052] The digital elevation model or real-scene 3D model of the inspection area can be obtained directly from the relevant platform based on the area range corresponding to the inspection area, or it can be constructed by collecting inspection images and combining the camera's attitude data and POS data at the time of shooting, using existing geographic model construction methods.
[0053] like Figure 2 The diagram shown is a system block diagram of a low-altitude forestry inspection target visual positioning system. This positioning system includes: The data acquisition module is configured to acquire inspection images of the inspection area, POS data and attitude data of the camera during image acquisition; The distortion detection module is configured to detect the degree of nonlinear distortion after the inspection image is projected onto the ground, and to set the mesh subdivision scale according to the degree of nonlinear distortion. The spherical subdivision module is configured to construct the field of view sphere of the camera during shooting based on shooting parameters, POS data and attitude data, and to perform spherical subdivision of the field of view sphere based on the grid subdivision scale to determine the spherical grid corresponding to the inspection image. The mapping module is configured to combine the camera's shooting parameters, POS data, and attitude data to determine the pixel-spherical mapping relationship between pixels and spherical grids in the inspection image, as well as the ground-spherical mapping relationship between ground coordinates and spherical grids within the inspection area. The target localization module is configured to identify and locate the pixel position of the forestry inspection target in the inspection image, and determine the ground coordinates corresponding to the forestry inspection target by combining the pixel-spherical mapping relationship and the ground-spherical mapping relationship.
[0054] Specifically, the positioning system includes a data acquisition module, a distortion detection module, a spherical partitioning module, a mapping module, and a target positioning module. The data acquisition module acquires inspection images of the inspection area captured by the camera, as well as POS data and attitude data of the inspection area captured by the camera. The distortion detection module detects the degree of nonlinear distortion in the projection of the inspection image onto the Earth's plane and adaptively sets the mesh partitioning scale for subsequent mesh generation based on the degree of nonlinear distortion.
[0055] The spherical meshing module constructs the field of view sphere when the camera captures inspection images, with the camera's optical center as the sphere center. Based on the set meshing scale, the field of view sphere is divided into meshes, and the spherical mesh corresponding to the inspection image is determined according to the shooting parameters, POS data and attitude data during shooting. This distributes the nonlinear distortion caused by projection evenly to each mesh, thereby reducing the deviation caused by projection.
[0056] The mapping module can determine the pixel-to-spherical mapping relationship between the pixels of the inspection image and the spherical grid of the field of view, as well as the ground-to-spherical mapping relationship between the ground coordinates of the inspection area and the spherical grid of the field of view, based on the camera's shooting parameters, POS data, and attitude data during the image capture. The target localization module can identify and locate the pixel positions of forestry inspection targets in the inspection image. Using the spherical grid as an intermediate reference, it maps the pixel positions of the forestry inspection targets in the inspection image to the spherical grid of the field of view through the pixel-to-spherical mapping relationship. Then, through the ground-to-spherical mapping relationship, it determines the ground grid corresponding to the forestry inspection target, and finally, it determines the ground coordinates of the forestry inspection target through the ground coordinates of the corresponding ground grid. This solves the problem of nonlinear distortion easily caused by traditional planar projection, improving the accuracy and reliability of the spatial geographical location of low-altitude forestry inspection targets.
[0057] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.
Claims
1. A visual positioning method for low-altitude forestry inspection targets, characterized in that, include: Acquire inspection images of the inspection area, as well as POS data and attitude data of the camera during image acquisition; The degree of nonlinear distortion of the inspection image projected onto the ground is detected, and the mesh subdivision scale is set according to the degree of nonlinear distortion. The field of view sphere constructed based on the camera optical center is divided according to the mesh subdivision scale, and the spherical mesh corresponding to the inspection image is determined according to the shooting parameters, POS data and attitude data. By combining the camera's shooting parameters, POS data, and attitude data, the pixel-spherical mapping relationship between pixels and spherical grids in the inspection image, as well as the ground-spherical mapping relationship between ground coordinates and spherical grids within the inspection area, are determined. Identify and locate the pixel position of the forestry inspection target in the inspection image, and determine the ground coordinates corresponding to the forestry inspection target by combining the pixel-spherical mapping relationship and the ground-spherical mapping relationship.
2. The visual positioning method for low-altitude forestry inspection targets according to claim 1, characterized in that, Detecting the degree of nonlinear distortion when the inspection image is projected onto the ground includes: The center pixel position of the forestry inspection target in the inspection image is located using a target detection algorithm; Based on the shooting parameters, locate the pixel position corresponding to the optical center of the camera, and combine it with the pixel position of the center point of the forestry inspection target to determine the degree of nonlinear distortion of the inspection image.
3. The low-altitude forestry inspection target visual positioning method according to claim 1, characterized in that, The meshing scale is set according to the degree of nonlinear distortion, including: The horizontal and vertical field of view of the camera are calculated based on the shooting parameters, and the horizontal and vertical angle steps are set according to the degree of nonlinear distortion. By combining the horizontal field of view, vertical field of view, horizontal angle step size, and vertical angle step size, the number of horizontal grids and the number of vertical grids are calculated, and the grid subdivision scale is determined.
4. The visual positioning method for low-altitude forestry inspection targets according to claim 1, characterized in that, The spherical grid corresponding to the inspection image is determined based on the shooting parameters, POS data, and attitude data, including: The horizontal and vertical field of view of the camera when capturing the inspection image are calculated based on the shooting parameters, and the field of view of the camera during shooting is determined by the horizontal and vertical field of view. The spherical grid corresponding to the inspection image in the field of view grid is determined by the field of view range, POS data, and attitude data.
5. The low-altitude forestry inspection target visual positioning method according to claim 1, characterized in that, Determine the pixel-to-sphere mapping relationship between pixels in the inspection image and the spherical grid, including: Based on the camera's shooting parameters, POS data, and attitude data, determine the pixel position of the camera's optical center in the inspection image, as well as the corresponding spherical grid in the field of view sphere. Based on the pixel position of the camera's optical center, the inspection image is divided into grids according to the grid subdivision scale, and the pixel grid corresponding to the camera's optical center is determined. Based on the pixel grid and spherical grid corresponding to the optical center of the camera, the mapping relationship between the pixel grid and the spherical grid is determined, and the pixel-spherical mapping relationship is obtained.
6. The visual positioning method for low-altitude forestry inspection targets according to claim 1, characterized in that, Determine the ground-spherical mapping relationship between the ground coordinates of the inspection area and the spherical grid, including: Based on the camera's shooting parameters, POS data, and attitude data, determine the ground point corresponding to the camera's optical center in the inspection area, as well as the corresponding spherical grid in the field of view grid sphere. Based on the ground point coordinates corresponding to the camera optical center, the ground in the inspection area is divided into grids according to the grid subdivision scale, and the ground grid corresponding to the camera optical center is determined. Based on the ground grid and spherical grid corresponding to the camera's optical center, the mapping relationship between the ground grid and the spherical grid is determined, thus obtaining the ground-spherical mapping relationship.
7. The visual positioning method for low-altitude forestry inspection targets according to claim 1, characterized in that, Identifying and locating the pixel positions of forestry inspection targets in the inspection images includes: The trained deep learning model identifies forestry inspection targets in the inspection images and locates the pixel positions of the forestry inspection targets.
8. The visual positioning method for low-altitude forestry inspection targets according to claim 1, characterized in that, Determine the ground coordinates corresponding to the forestry inspection targets, including: Based on the pixel position of the forestry inspection target, the spherical grid corresponding to the forestry inspection target is determined through the pixel-spherical mapping relationship; Based on the spherical grid corresponding to the forestry inspection target, the ground grid corresponding to the forestry inspection target is determined through the ground-spherical mapping relationship, and the ground coordinates of the forestry inspection target are located.
9. The visual positioning method for low-altitude forestry inspection targets according to claim 1, characterized in that, Also includes: Obtain a real-world 3D model of the inspection area, and determine the actual elevation of the forestry inspection target by combining the ground coordinates of the target.
10. A low-altitude forestry inspection target visual positioning system, characterized in that, include: The data acquisition module is configured to acquire inspection images of the inspection area, POS data and attitude data of the camera during image acquisition; The distortion detection module is configured to detect the degree of nonlinear distortion after the inspection image is projected onto the ground, and to set the mesh subdivision scale according to the degree of nonlinear distortion. The spherical subdivision module is configured to construct the field of view sphere of the camera during shooting based on shooting parameters, POS data and attitude data, and to perform spherical subdivision of the field of view sphere based on the grid subdivision scale to determine the spherical grid corresponding to the inspection image. The mapping module is configured to combine the camera's shooting parameters, POS data, and attitude data to determine the pixel-spherical mapping relationship between pixels and spherical grids in the inspection image, as well as the ground-spherical mapping relationship between ground coordinates and spherical grids within the inspection area. The target localization module is configured to identify and locate the pixel position of the forestry inspection target in the inspection image, and determine the ground coordinates corresponding to the forestry inspection target by combining the pixel-spherical mapping relationship and the ground-spherical mapping relationship.