Large-scene UAV positioning method and system based on multi-source cross-modal image layered matching

Abstract

The present invention discloses a large-scale drone positioning method and system using multi-source cross-modal image layered matching. The method comprises generating an image pyramid with different ground resolutions and presented as a tile map based on a satellite optical map of the drone airport area, obtaining a real-shot image of the drone airport area and correcting it to have the same orientation as the satellite optical map; determining a tile map with the same ground target based on the ground resolution of the real-shot image; splicing the found tile map with its neighborhood tile map to obtain a reference image; matching the real-shot image and the reference image using a layered matching method, correcting the geographic coordinates of the feature points in the real-shot image based on the corresponding feature points, and performing pose calculation to solve the drone's pose. The present invention aims to achieve visually assisted positioning of drones, so that when the positioning system can only provide low-precision positions, the drone can be accurately positioned in large scenes through visual assistance.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a large-scene UAV positioning method and system based on multi-source cross-modal image layered matching. Background Art

[0002] Providing accurate navigation and positioning for drones is crucial for drone operations.

[0003] For traditional drone positioning, the Global Navigation Satellite System (GNSS) and the Inertial Navigation System (INS) are commonly used. GNSS relies on communication between satellites and ground-based receivers to determine the position and velocity of the vehicle. It offers advantages such as portability, low cost, and the ability to provide real-time positioning, making it the most widely used navigation system. However, in complex environments, drones are subject to electromagnetic interference and shielding during flight, and the external environment also significantly impacts GNSS. This makes it difficult to provide accurate positioning for drones, reducing their survival rate and mission completion rate. INS relies on time-integrated angular and linear accelerations provided by gyroscopes and accelerometers to determine velocity and displacement. It has the advantages of being immune to electromagnetic interference, compact, and inexpensive. However, due to its positioning characteristics, its errors can grow over time, making it incapable of providing long-term, high-precision positioning information.

[0004] Therefore, how to provide accurate navigation information for drones through auxiliary positioning means and overcome electromagnetic interference, shielding and environmental influences has become a key technical problem that needs to be solved in drone positioning technology. With the development of computer vision technology, visual positioning technology has been booming. On the basis of obtaining the precise geographical location of the pixels in the real-shot image, existing methods have provided a method of using PnP (Perspective-n-Points Problem) pose estimation to solve the carrier pose. Since visual positioning technology is a fully autonomous positioning technology, it is not affected by electromagnetic interference, and the error does not diverge over time, it has the advantage of working all day and all weather. However, there are errors in the geographical location of the pixels in the real-shot image obtained by the existing technology, which makes the visual positioning technology have the problem of insufficient positioning accuracy. Summary of the Invention

[0005] Technical problem to be solved by the present invention: In response to the above-mentioned problems of the prior art, a large-scene UAV positioning method and system with multi-source cross-modal image layered matching are provided. The present invention aims to realize visually assisted positioning of UAVs, so as to achieve precise positioning of UAVs in large scenes through visual assistance when the positioning system can only provide low-precision positions.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] A large-scene UAV localization method based on multi-source cross-modal image layered matching includes:

[0008] S1: Generate an image pyramid with different ground resolutions and presented as a tile map based on the satellite optical map of the drone airport area, obtain the actual image of the drone airport area and calibrate it to the same direction as the satellite optical map;

[0009] S2: Determine the corresponding tile map level based on the ground resolution of the real-shot image, and search the tile map corresponding to the tile map level based on the geographic coordinates of the central pixel of the real-shot image to find a tile map with the same ground object as the real-shot image;

[0010] S3, stitching the found tile map with its neighboring tile maps to obtain a reference map;

[0011] S4, performing image matching between the real image and the reference image using a layered matching method to obtain corresponding feature points and pixel transformation matrices between the real image and the reference image;

[0012] S5, correcting the geographic coordinates of the feature points with position errors in the real image based on the precise geographic coordinates of the reference image and the corresponding feature points in the real image to obtain geographic coordinates with higher accuracy;

[0013] S6, performing posture calculation to solve the UAV posture according to the corrected geographic coordinates of the feature points in the real-shot image.

[0014] Optionally, determining the corresponding tile map level according to the ground resolution of the real-shot image in step S2 includes:

[0015] S2.1, calculate the distance between the real image and each tile map according to the following formula d ;

[0016] ,

[0017] In the above formula, and are the two-axis ground resolutions of the real-shot images, and are the two-axis ground resolutions of the i-th tile map;

[0018] S2.2, find the distance d The tile map level corresponding to the smallest tile map is used as the tile map level corresponding to the ground resolution of the real-shot image.

[0019] Optionally, in step S2, when searching the tile map corresponding to the tile map level according to the geographic coordinates of the center pixel point of the real-shot image to find a tile map with the same ground object as the real-shot image, the function expression of the constraint condition satisfied by the tile map with the same ground object as the real-shot image is:

[0020] ,

[0021] In the above formula, is the geographic coordinate of the center pixel of the real-shot image, is the coordinate of the upper left corner of the i-th tile map, is the coordinate of the lower right corner of the i-th tile map.

[0022] Optionally, performing image matching between the real image and the reference image using a layered matching method in step S4 includes:

[0023] S4.1, downsample the real image and the reference image respectively, and then use the preset matching algorithm to match the downsampled reference image and the infrared map to obtain a set of feature point pairs ,in is the kth feature point of the reference image, is the kth feature point of the real image, is the total number of feature point pairs;

[0024] S4.2, clustering the feature points of the real image and the reference image to obtain the coordinates of the cluster centers;

[0025] S4.3, respectively restore the cluster center coordinates to the original ground resolution image coordinates, and crop a neighborhood of a specified size as the region of interest according to the original ground resolution image coordinates to obtain a matching image of the region of interest;

[0026] S4.4, match the real image and the reference image of the region of interest using a preset matching algorithm to obtain a set of feature point pairs ,in Match the kth feature point of the image to the region of interest of the reference image, Match the kth feature point of the image to the region of interest of the real image, is the total number of feature point pairs.

[0027] Optionally, when downsampling the real image and the reference image in step S4.1, the reference image is downsampled. Multiple downsampling, downsampling is performed on the real-shot image according to the following formula:

[0028] ,

[0029] In the above formula, and is the downsampling ratio of the x and y axis directions of the real image, and is the two-axis ground resolution of the reference image, and is the two-axis ground resolution of the real image, The downsampling factor to perform on the reference image.

[0030] Optionally, in step S4.2, the function expression for obtaining the cluster center coordinates by clustering the feature points of the real image and the reference image is:

[0031] , ,

[0032] In the above formula, is the cluster center coordinate of the real image, and are the x-coordinate and y-coordinate of the i-th feature point in the real image, are the cluster center coordinates of the reference graph, and are the x-coordinate and y-coordinate of the i-th feature point of the reference image, is the total number of feature point pairs.

[0033] Optionally, in step S4.3, the function expression for restoring the cluster center coordinates to the original ground resolution image coordinates is:

[0034] , ,

[0035] In the above formula, is the cluster center coordinates after the real-shot image is restored to the original ground resolution image coordinates, and is the downsampling ratio of the x and y axis directions of the real image, is the cluster center coordinate after the reference image is restored to the original ground resolution image coordinate, The downsampling factor performed on the reference image, is the cluster center coordinate of the real image, are the cluster center coordinates of the reference graph.

[0036] In addition, the present invention also provides a large-scene UAV positioning system with multi-source cross-modal image layered matching, comprising a microprocessor and a memory connected to each other, wherein the microprocessor is programmed or configured to execute the large-scene UAV positioning method with multi-source cross-modal image layered matching.

[0037] In addition, the present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program is used to be programmed or configured by a microprocessor to execute the large-scene drone positioning method of multi-source cross-modal image layered matching.

[0038] In addition, the present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the large-scene drone positioning method of multi-source cross-modal image layered matching through a processor.

[0039] Compared with the existing technology, the present invention has the following main advantages: the large-scene UAV positioning method of the present invention using multi-source cross-modal image layered matching can achieve precise positioning of the UAV in large scenes through visual assistance when the main positioning system of the UAV (such as GNSS) can only provide low-precision positioning. The satellite optical map reference image area is determined by the low-precision position provided by the main positioning system, and then the reference image and the actual image are layered image matched to obtain the precise geographic location coordinates of the feature points in the actual image. Finally, the UAV posture is solved and is less affected by electromagnetic interference and the external environment, and has high positioning accuracy, which can provide navigation guarantee for the UAV to perform tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 Schematic diagram of the basic process of the method of the embodiment of the present invention.

[0042] Figure 2 Schematic diagram of tile maps at different levels in an embodiment of the present invention, where (a) and (b) are a group of tile maps at different resolutions on the 12th layer, and (c) and (d) are a group of tile maps at different resolutions on the 14th layer.

[0043] Figure 3 Schematic diagram of reference image and infrared image pairs in an embodiment of the present invention, where (a) and (b) are the first set of reference image and infrared image pairs, (a) is the reference image, and (b) is the actual image; (c) and (d) are the second set of reference image and infrared image pairs, (c) is the reference image, and (d) is the actual image.

[0044] Figure 4Schematic diagram of the results of using hierarchical matching between a reference image and an infrared image in an embodiment of the present invention, where (a) is the result of coarse matching, (b) is the result of fine matching, and (c) is the final result. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0046] like Figure 1 As shown, the large-scene UAV positioning method based on multi-source cross-modal image layered matching in this embodiment includes:

[0047] S1: Generate an image pyramid with different ground resolutions and presented as a tile map based on the satellite optical map of the drone airport area, obtain the actual image of the drone airport area and calibrate it to the same direction as the satellite optical map;

[0048] S2: Determine the corresponding tile map level based on the ground resolution of the real-shot image, and search the tile map corresponding to the tile map level based on the geographic coordinates of the central pixel of the real-shot image to find a tile map with the same ground object as the real-shot image;

[0049] S3, stitching the found tile map with its neighboring tile maps to obtain a reference map;

[0050] S4, performing image matching between the real image and the reference image using a layered matching method to obtain corresponding feature points and pixel transformation matrices between the real image and the reference image;

[0051] S5, correcting the geographic coordinates of the feature points with position errors in the real image based on the precise geographic coordinates of the reference image and the corresponding feature points in the real image to obtain geographic coordinates with higher accuracy;

[0052] S6, performing posture calculation to solve the UAV posture according to the corrected geographic coordinates of the feature points in the real-shot image.

[0053] In this embodiment, step S1 is an image preprocessing step. In step S1, the human-machine airport area can be divided according to the flight speed and duration of the drone, generally within plus or minus 20 degrees of longitude and latitude. After generating an image pyramid with different ground resolutions and presented in the form of a tile map based on the satellite optical map (including geographic location information) of the drone airport area, the size of each tile map is , and the geographic coordinates of the pixel in the upper left corner of the tile map are known (WGS84 coordinate system). The image pyramid generally has 18 layers, that is, it includes 18 types of ground resolutions, with the highest being about 0.5m. Figure 2 , showing examples of tile maps at different levels, where (a) and (b) are a set of tile maps of different resolutions at the 12th layer, and (c) and (d) are a set of tile maps of different resolutions at the 14th layer. The real-shot image used in this embodiment is a real-shot infrared image, the purpose of which is to realize drone positioning at night. If only daytime conditions are considered, ordinary visible light real-shot images can be used. Based on the flight direction of the drone, the real-shot infrared image is corrected to due north (all satellite optical maps are due north) to reduce the difficulty of image matching. The flight direction of the drone is known, which is also the direction of the infrared image, and the real-shot image can be rotated to due north.

[0054] In step S2 of this embodiment, determining the corresponding tile map level according to the ground resolution of the real-shot image includes:

[0055] S2.1, calculate the distance between the real image and each tile map according to the following formula d ;

[0056] ,

[0057] In the above formula, and are the two-axis ground resolutions of the real-shot images, and are the two-axis ground resolutions of the i-th tile map;

[0058] S2.2, find the distance d The tile map level corresponding to the smallest tile map is used as the tile map level corresponding to the ground resolution of the real-shot image.

[0059] In step S2 of this embodiment, the corresponding tile map level is determined based on the ground resolution of the real-shot image by first determining the tile map level with the smallest resolution difference based on the ground resolution of the infrared map, and then performing a geographic match. This can reduce the scale difference between the reference image and the real-shot image, making subsequent image matching easier. In step S2 of this embodiment, when searching the tile map corresponding to the tile map level based on the geographic coordinates of the central pixel point of the real-shot image to find a tile map with the same ground target as the real-shot image, the function expression for the constraint condition satisfied by the tile map with the same ground target as the real-shot image is:

[0060] ,

[0061] In the above formula, is the geographic coordinate of the center pixel of the real-shot image, is the coordinate of the upper left corner of the i-th tile map, is the coordinate of the lower right corner of the i-th tile map. The geographic coordinate of the upper left corner of the infrared map is known. , two-axis ground resolution and , pixel resolution , then the center point coordinates That is:

[0062] , ,

[0063] Similarly, it is known that the coordinates of the upper left corner of each tile map are , two-axis ground resolution and , pixel resolution ,in Represents the tile map serial number, lower right corner coordinates That is:

[0064] , .

[0065] In step S3 of this embodiment, when the tile map found is spliced ​​with its neighborhood tile map to obtain a reference map, the tile maps found have similar ground resolutions, but their pixel resolutions are unified as , while the pixel resolution of the real infrared map is generally , which results in a small overlap between the tile map and the infrared image, making it difficult to match successfully. To address this situation, step S3 of this embodiment expands the tile map and splices the tile maps of its neighborhood. The number of splices can be specified as needed. For example, as an optional implementation, this embodiment uses the neighborhood The tile maps are stitched together to get A reference map of the size (with precise geographic coordinates) is used for subsequent image matching. Figure 3 Examples of reference image and infrared image pairs after geographic matching and tile stitching are shown, where (a) and (b) are the first set of reference image and infrared image pairs, (a) is the reference image, and (b) is the real image; (c) and (d) are the second set of reference image and infrared image pairs, (c) is the reference image, and (d) is the real image.

[0066] like Figure 1 As shown, in step S4, the real image and the reference image are matched using a layered matching method, including:

[0067] S4.1, coarse matching: downsample the real image and the reference image respectively, and then use the preset matching algorithm to match the downsampled reference image and the infrared map to obtain a set of feature point pairs. ,in is the kth feature point of the reference image, is the kth feature point of the real image, is the total number of feature point pairs;

[0068] S4.2, feature point clustering: cluster the feature points of the real image and the reference image respectively to obtain the cluster center coordinates;

[0069] S4.3, ROI cropping: restore the cluster center coordinates to the original ground resolution image coordinates, and crop a neighborhood of a specified size as the ROI based on the original ground resolution image coordinates to obtain the ROI matching image;

[0070] S4.4, precise matching: Use the preset matching algorithm to match the real image and the reference image of the region of interest to obtain a set of feature point pairs. ,in Match the kth feature point of the image to the region of interest of the reference image, Match the kth feature point of the image to the region of interest of the real image, is the total number of feature point pairs.

[0071] It should be noted that the preset matching algorithm can adopt various known matching algorithms as needed. This embodiment involves cross-modal remote sensing image matching. However, deep learning-based methods are limited by the dataset and have difficulty solving cross-modal remote sensing image matching. Traditional mathematical matching methods have difficulty in matching successfully due to the high image resolution of the reference image and the infrared map. Therefore, the MSFT algorithm (see: Maoqing Hu, Bin Sun, Xudong Kang, Shutao Li, Multiscale structural feature transform for multi-modal image matching, Information Fusion, Vol 95, 2023, pp 341-354.) is adopted as the preset matching algorithm.

[0072] In step S4.1 of this embodiment, when downsampling the real image and the reference image, the reference image is downsampled. Multiple downsampling, downsampling is performed on the real-shot image according to the following formula:

[0073] ,

[0074] In the above formula, and is the downsampling ratio of the x and y axis directions of the real image, and is the two-axis ground resolution of the reference image, and is the two-axis ground resolution of the real image, The downsampling multiple of the reference image can be set as needed. For example, as an optional implementation, in this embodiment, the downsampling multiple of the reference image is set to 4, then:

[0075] ,

[0076] Then, the MSFT algorithm is used to match the downsampled reference image and the infrared map to obtain a set of feature point pairs.

[0077] In step S4.2 of this embodiment, the function expression for clustering the feature points of the real image and the reference image to obtain the cluster center coordinates is:

[0078] , ,

[0079] In the above formula, is the cluster center coordinate of the real image, and are the x-coordinate and y-coordinate of the i-th feature point in the real image, are the cluster center coordinates of the reference graph, and are the x-coordinate and y-coordinate of the i-th feature point of the reference image, is the total number of feature point pairs.

[0080] In step S4.3 of this embodiment, the function expression for restoring the cluster center coordinates to the original ground resolution image coordinates is:

[0081] ,

[0082] ,

[0083] In the above formula, is the cluster center coordinates after the real-shot image is restored to the original ground resolution image coordinates, and is the downsampling ratio of the x and y axis directions of the real image, is the cluster center coordinate after the reference image is restored to the original ground resolution image coordinate, The downsampling factor performed on the reference image, is the cluster center coordinate of the real image, is the cluster center coordinate of the reference image. In this embodiment, based on the cluster center coordinates obtained by the above formula, the corresponding neighborhoods are clipped to obtain the area of ​​interest, and the clipping size is , which are matching image pairs that have the same ground targets and are easy to match.

[0084] In the precise matching in step S4.4 of this embodiment, the real image and the reference image of the region of interest are matched using the MSFT algorithm to obtain a set of feature point pairs. ,in Match the kth feature point of the image to the region of interest of the reference image, Match the kth feature point of the image to the region of interest of the real image, is the total number of feature point pairs, Figure 4 A comparison chart of coarse matching, fine matching and final results is shown, where (a) is the result of coarse matching, (b) is the result of fine matching, and (c) is the final result.

[0085] Then, in step S5, the geographic coordinates of the feature points with position errors in the real-shot image are corrected according to the precise geographic coordinates of the reference image and the corresponding feature points in the real-shot image to obtain more precise geographic coordinates; in step S6, the pose is solved to obtain the pose of the drone according to the corrected geographic coordinates of the feature points in the real-shot image. In this embodiment, the pose of the drone is solved using the PnP pose solution method. The PnP pose solution method is a well-known pose solution method, which includes the precise geographic coordinates of the feature points obtained in step S5. and the corresponding pixel coordinates First, use P3P to solve the initial pose, and then use Bundle Adjustment (BA) to iteratively optimize the pose to obtain the final accurate pose. The principle of P3P solving the initial pose is as follows: given three pairs of matching pairs of known geographic coordinates and pixel coordinates, that is, the pixel coordinates and geographic coordinates of the infrared image feature points obtained in step five. Then, based on similar triangles, a set of quadratic equations is constructed jointly, and then the Wu elimination method is used to solve it to obtain the 3D coordinates of these three pairs of feature points in the camera coordinate system. Finally, based on the 3D-3D point pair (camera coordinate system-geographic coordinate system), the pose of the camera in the geographic coordinate system is estimated. For specific calculation methods, see the reference (Gao Xiang and Zhang Tao, "Fourteen Lectures on Visual SLAM," Publishing House of Electronics Industry, 2017). The principle of Bundle Adjustment (BA) involves constructing an optimization function based on the initial pose provided by the PnP algorithm, with the reprojection error as the objective function and the UAV pose as the optimization term. Then, the UAV pose is iteratively optimized using methods such as Gauss-Newton iteration to obtain the final accurate pose. For specific calculation methods, see the reference (Gao Xiang and Zhang Tao, "Fourteen Lectures on Visual SLAM," Publishing House of Electronics Industry, 2017).

[0086] In summary, the large-scene UAV positioning method of multi-source cross-modal image layered matching in this embodiment includes geographic matching based on the UAV aerial infrared image and the satellite optical map to obtain a reference map with precise geographic coordinates near the UAV; by performing layered image matching on the infrared image and the reference map to obtain pixel coordinate correspondence between the two, wherein the layered image matching algorithm mainly includes the steps of coarse matching, clustering, cropping, and fine matching; finally, based on the pixel coordinate correspondence between the infrared image and the reference map, the geographic coordinates of the infrared image are corrected, and finally the precise positioning of the UAV is achieved through pose solution. The present invention can achieve precise positioning of the UAV in large scenes through visual assistance when the UAV's global positioning system is interfered with and can only provide low-precision positions.

[0087] In addition, this embodiment also provides a large-scene drone positioning system using multi-source cross-modal image layered matching, comprising an interconnected microprocessor and a memory, wherein the microprocessor is programmed or configured to execute the large-scene drone positioning method using multi-source cross-modal image layered matching. This embodiment also provides a computer-readable storage medium storing a computer program, wherein the computer program is programmed or configured by the microprocessor to execute the large-scene drone positioning method using multi-source cross-modal image layered matching. This embodiment also provides a computer program product, comprising a computer program or instructions, wherein the computer program or instructions are programmed or configured to execute the large-scene drone positioning method using multi-source cross-modal image layered matching through a processor.

[0088] Those skilled in the art should understand that the technical solution provided by the present invention may be in the form of a method, a system, or a computer program product. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the functions described in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including the instruction device, which implements the function specified in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0089] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A large-scene UAV positioning method based on multi-source cross-modal image layered matching, characterized by: include: S1: Generate an image pyramid with different ground resolutions and presented as a tile map based on the satellite optical map of the drone airport area, obtain the actual image of the drone airport area and calibrate it to the same direction as the satellite optical map; S2: Determine the corresponding tile map level based on the ground resolution of the real-shot image, and search the tile map corresponding to the tile map level based on the geographic coordinates of the central pixel of the real-shot image to find a tile map with the same ground object as the real-shot image; S3, stitching the found tile map with its neighboring tile maps to obtain a reference map; S4, performing image matching between the real image and the reference image using a layered matching method to obtain corresponding feature points and pixel transformation matrices between the real image and the reference image; S5, correcting the geographic coordinates of the feature points with position errors in the real image based on the precise geographic coordinates of the reference image and the corresponding feature points in the real image to obtain geographic coordinates with higher accuracy; S6, performing posture calculation to solve the UAV posture according to the corrected geographic coordinates of the feature points in the real-shot image.

2. The large-scene UAV positioning method based on multi-source cross-modal image layered matching according to claim 1 is characterized in that: Determining the corresponding tile map level according to the ground resolution of the real-shot image in step S2 includes: S2.1, calculate the distance between the real image and each tile map according to the following formula d ; ， In the above formula, and are the two-axis ground resolutions of the real-shot images, and are the two-axis ground resolutions of the i-th tile map; S2.2, find the distance d The tile map level corresponding to the smallest tile map is used as the tile map level corresponding to the ground resolution of the real-shot image.

3. The large-scene UAV positioning method based on multi-source cross-modal image layered matching according to claim 1 is characterized in that: In step S2, when searching the tile map corresponding to the tile map level according to the geographic coordinates of the center pixel point of the real-shot image to find a tile map with the same ground object as the real-shot image, the function expression of the constraint condition satisfied by the tile map with the same ground object as the real-shot image is: ， In the above formula, is the geographic coordinate of the center pixel of the real-shot image, is the coordinate of the upper left corner of the i-th tile map, is the coordinate of the lower right corner of the i-th tile map.

4. The large-scene UAV positioning method based on multi-source cross-modal image layered matching according to claim 1 is characterized in that: In step S4, the image matching between the real image and the reference image using a layered matching method includes: S4.1, downsample the real image and the reference image respectively, and then use the preset matching algorithm to match the downsampled reference image and the infrared map to obtain a set of feature point pairs ,in is the kth feature point of the reference image, is the kth feature point of the real image, is the total number of feature point pairs; S4.2, clustering the feature points of the real image and the reference image to obtain the coordinates of the cluster centers; S4.3, respectively restore the cluster center coordinates to the original ground resolution image coordinates, and crop a neighborhood of a specified size as the region of interest according to the original ground resolution image coordinates to obtain a matching image of the region of interest; S4.4, match the real image and the reference image of the region of interest using a preset matching algorithm to obtain a set of feature point pairs ,in Match the kth feature point of the image to the region of interest of the reference image, Match the kth feature point of the image to the region of interest of the real image, is the total number of feature point pairs.

5. The large-scene UAV positioning method based on multi-source cross-modal image layered matching according to claim 4 is characterized in that: When downsampling the real image and the reference image in step S4.1, the reference image is downsampled. Multiple downsampling, downsampling is performed on the real-shot image according to the following formula: ， In the above formula, and is the downsampling ratio of the x and y axis directions of the real image, and is the two-axis ground resolution of the reference image, and is the two-axis ground resolution of the real image, The downsampling factor to perform on the reference image.

6. The large-scene UAV positioning method based on multi-source cross-modal image layered matching according to claim 4 is characterized in that: In step S4.2, the function expression for clustering the feature points of the real image and the reference image to obtain the cluster center coordinates is: ， ， In the above formula, is the cluster center coordinate of the real image, and are the x-coordinate and y-coordinate of the i-th feature point in the real image, are the cluster center coordinates of the reference graph, and are the x-coordinate and y-coordinate of the i-th feature point of the reference image, is the total number of feature point pairs.

7. The large-scene UAV positioning method based on multi-source cross-modal image layered matching according to claim 4 is characterized in that: The function expressions for restoring the cluster center coordinates to the original ground resolution image coordinates in step S4.3 are: ， ， In the above formula, is the cluster center coordinates after the real-shot image is restored to the original ground resolution image coordinates, and is the downsampling ratio of the x and y axis directions of the real image, is the cluster center coordinate after the reference image is restored to the original ground resolution image coordinate, The downsampling factor performed on the reference image, is the cluster center coordinate of the real image, are the cluster center coordinates of the reference graph.

8. A large-scene UAV positioning system with multi-source cross-modal image hierarchical matching, comprising an interconnected microprocessor and memory, characterized in that: The microprocessor is programmed or configured to execute the large-scene UAV positioning method based on multi-source cross-modal image layered matching as described in any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, wherein: The computer program is used to be programmed or configured by a microprocessor to execute the large-scene UAV positioning method based on multi-source cross-modal image layered matching as described in any one of claims 1 to 7.

10. A computer program product comprising a computer program or instructions, characterized in that The computer program or instruction is programmed or configured to execute the large-scene UAV positioning method based on multi-source cross-modal image layered matching as described in any one of claims 1 to 7 through a processor.

Images