Unmanned aerial vehicle rapid and accurate positioning method and device based on satellite image matching

Abstract

The invention relates to the technical field of image feature matching of computer vision, in particular to an unmanned aerial vehicle quick and accurate positioning method and device based on satellite image matching, and the method comprises the steps: obtaining unmanned aerial vehicle route planning information, unmanned aerial vehicle data flow and reference satellite image data; key frames and non-key frames in the unmanned aerial vehicle data flow are retrieved according to the unmanned aerial vehicle route planning information; performing different-source image matching according to the key frame and the reference satellite image data, executing single-frame absolute positioning according to a different-source image matching result to obtain a single-frame absolute positioning result, and executing inter-frame relative positioning according to the single-frame absolute positioning result and the non-key frame to obtain the position of the unmanned aerial vehicle. Therefore, the problems that high precision and high processing speed are difficult to meet at the same time in the prior art, the matching precision and robustness can be remarkably reduced when an image with a large rotation angle is processed, and the influence of illumination change, view angle change, resolution difference and the like on the positioning accuracy is large are solved.

Description

Technical Field

[0001] The present application relates to the technical field of image feature matching of computer vision, and in particular to a method and device for quickly and accurately positioning a drone based on satellite image matching. Background Art

[0002] With the rapid development and widespread application of drone technology, drones are playing an increasingly important role in agricultural monitoring, environmental monitoring, urban planning, emergency rescue and other fields. However, in practical applications, the precise positioning of drones is a key issue, especially in the absence of satellite signals (such as GNSS signals). Traditional drone positioning methods mainly rely on satellite navigation systems, but in some environments, such as indoors, forests, and urban canyons, satellite signals may be severely interfered or completely unavailable, resulting in the inability to achieve high-precision positioning.

[0003] To solve this problem, vision-based positioning methods have gradually attracted attention. Visual positioning methods analyze the image data taken by drones and combine them with pre-built high-resolution base map data to achieve accurate positioning of drones. However, existing visual positioning methods still face many challenges in practical applications:

[0004] 1. High-precision positioning usually requires complex calculations and a large amount of data processing, resulting in slow processing speeds and difficulty meeting the needs of real-time positioning. Although fast positioning methods have fast processing speeds, their accuracy is often not high enough and it is difficult to meet the requirements of high-precision positioning.

[0005] 2. When the drone is in flight, the captured images may rotate greatly due to the change of posture. When the existing matching algorithm processes images with large rotation angles, the matching accuracy and robustness will be significantly reduced, affecting the positioning effect.

[0006] 3. Images taken by drones often have problems such as lighting changes, perspective changes, and resolution differences. These problems will increase the difficulty of matching and affect the accuracy of positioning. Summary of the invention

[0007] The present application provides a method and device for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching, in order to solve the problems that related technologies are difficult to simultaneously meet high precision and fast processing speed, matching accuracy and robustness will significantly decrease when processing images with large rotation angles, and changes in lighting, perspective and resolution have a significant impact on positioning accuracy.

[0008] The first aspect of the present application provides a method for rapid and accurate positioning of a UAV based on satellite image matching, comprising the following steps: obtaining UAV route planning information, UAV data stream and reference satellite image data; retrieving key frames and non-key frames in the UAV data stream according to the UAV route planning information, wherein the key frames are frames determined by the target posture information in the UAV route planning information, and the non-key frames are frames other than the key frames in the UAV data stream; performing heterogeneous image matching according to the key frames and the reference satellite image data, performing single-frame absolute positioning according to the heterogeneous image matching results to obtain a single-frame absolute positioning result, and performing relative positioning between frames according to the single-frame absolute positioning result and the non-key frames to obtain the UAV position.

[0009] Optionally, the key frame includes at least one of an image taken when the drone's posture changes, an image with target texture information in an area corresponding to the reference satellite image data, an image taken when the drone takes off, and an image taken at a preset interval.

[0010] Optionally, heterogeneous image matching is performed based on the key frame and the reference satellite image data, including: identifying the base map in the reference satellite image data; respectively extracting the first multi-level features of the base map and the second multi-level features of the key frame; performing feature transformation on the first multi-level features based on the intra-domain attention mechanism to obtain a first feature map, and performing feature transformation on the second multi-level features based on the inter-domain attention mechanism to obtain a second feature map; performing coarse-level matching on the first feature map and the second feature map to obtain a coarse-level matching result, and performing fine-level matching on the coarse-level matching result to obtain a fine-level matching result.

[0011] Optionally, the prediction formula for coarse-level matching is:

[0012]

[0013] in, is a rough matching; MNN(﹒) is a mutual nearest neighbor criterion algorithm; P c is the confidence matrix; θ C is the confidence threshold.

[0014] Optionally, performing fine-level matching on the coarse-level matching result to obtain a fine-level matching result includes: locating the coarse-level matching result In fine-level feature maps and The position of the local window is cropped, and the cropped features in each window are transformed multiple times to generate two transformations. and The local feature map centered on and Will The center vector of All vectors in are associated to obtain a heat map, where the heat map represents Each pixel in the neighborhood of The matching probability is obtained by calculating the expectation on the probability distribution. B The final matching position with sub-pixel accuracy on Collect all matches Generate the final fine-level match M f .

[0015] Optionally, performing single-frame absolute positioning according to the heterogeneous image matching result to obtain the single-frame absolute positioning result includes: obtaining the same-name control points according to the object coordinate transformation and the image coordinate transformation of the heterogeneous image matching result; and calculating the position of the UAV according to the same-name control points, wherein the calculation formula of the position of the UAV is:

[0016]

[0017] Among them, (x0, y0) is the image principal point, (x, y) is the image point coordinate, f is the principal distance, (X A ,Y A ,Z A ) is the object coordinate, (X S ,Y S ,Z S ) is the three-dimensional coordinate of the camera imaging center in the object coordinate system; a1a2······c3 are obtained from the posture information.

[0018] Optionally, the formula for object coordinate transformation is:

[0019] X=A·l+B·r+C

[0020] Y=D·l+E·r+F

[0021] Among them, A is the horizontal component of the pixel size in the X direction; B is the vertical component of the pixel size in the X direction (usually 0); C is the X coordinate of the upper left corner of the image; D is the vertical component of the pixel size in the Y direction (usually 0); E is the horizontal component of the pixel size in the Y direction; F is the Y coordinate of the upper left corner of the image; (l, r) is the pixel coordinate; (X, Y) is the geographic coordinate; the formula for image coordinate transformation is:

[0022]

[0023] Among them, dx, dy represents the physical size of a pixel; (u0, v0) represents the position of the image principal point in the image plane coordinate system; (u, v) is the drone image coordinate; (x, y) is the image plane coordinate.

[0024] Optionally, the matching method used for relative positioning between frames includes FAST corner point extraction, BRIFE descriptor calculation, Hamming distance-based and verified epipolar constraints.

[0025] The second aspect of the present application provides a device for rapid and precise positioning of a UAV based on satellite image matching, including: an acquisition module, used to acquire UAV route planning information, UAV data stream and reference satellite image data; a retrieval module, used to retrieve key frames and non-key frames in the UAV data stream according to the UAV route planning information, wherein the key frames are frames determined by the target posture information in the UAV route planning information, and the non-key frames are frames other than the key frames in the UAV data stream; an execution module, used to perform heterogeneous image matching according to the key frames and the reference satellite image data, perform single-frame absolute positioning according to the heterogeneous image matching results to obtain single-frame absolute positioning results, and perform inter-frame relative positioning according to the single-frame absolute positioning results and non-key frames to obtain the UAV position.

[0026] The third aspect of the present application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the first aspect of the method for rapid and precise positioning of a drone based on satellite image matching.

[0027] Therefore, this application includes the following beneficial effects:

[0028] The embodiment of the present application obtains the UAV route planning information, the UAV data stream and the reference satellite image data, retrieves the key frames and non-key frames in the UAV data stream, performs heterogeneous image matching according to the key frames and the reference satellite image data, performs single-frame absolute positioning according to the heterogeneous image matching result to obtain the single-frame absolute positioning result, performs inter-frame relative positioning according to the single-frame absolute positioning result and the non-key frames to obtain the UAV position. The whole process is short in time and high in precision. The heterogeneous image matching algorithm with rotation invariance is adopted to ensure the matching accuracy and robustness of images with large rotation angles, and ensure the accuracy of positioning when the illumination changes, the viewing angle changes and the resolution difference. Thus, it solves the problems that the related technology is difficult to meet the high precision and fast processing speed at the same time, the matching accuracy and robustness will be significantly reduced when processing images with large rotation angles, and the illumination changes, the viewing angle changes and the resolution difference have a great influence on the accuracy of positioning.

[0029] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0031] Figure 1 A flowchart of a method for rapid and accurate positioning of a drone based on satellite image matching provided according to an embodiment of the present application;

[0032] Figure 2 The overall technical flow chart of a remote sensing image interest point matching method based on a twin encoding spatial attention mechanism provided according to an embodiment of the present application;

[0033] Figure 3 A diagram of an overall algorithm structure provided according to an embodiment of the present application;

[0034] Figure 4 A diagram showing an epipolar constraint provided according to an embodiment of the present application;

[0035] Figure 5 A diagram showing calculation results of other algorithms provided according to an embodiment of the present application;

[0036] Figure 6 A diagram showing calculation results of a heterogeneous image matching algorithm with rotation invariance provided according to an embodiment of the present application;

[0037] Figure 7 This is an example diagram of a device for rapid and accurate positioning of a drone based on satellite image matching provided according to an embodiment of the present application;

[0038] Figure 8 It is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0039] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0040] The following describes the method and device for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching in the embodiment of the present application with reference to the accompanying drawings. In view of the fact that the related technologies mentioned in the above background technology are difficult to simultaneously meet high precision and fast processing speed, the matching accuracy and robustness will be significantly reduced when processing images with large rotation angles, and the illumination changes, perspective changes and resolution differences have a greater impact on the accuracy of positioning, the present application provides a method for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching, in which the route planning information of the unmanned aerial vehicle, the unmanned aerial vehicle data stream and the reference satellite image data are obtained, the key frames and non-key frames in the unmanned aerial vehicle data stream are retrieved, and heterogeneous image matching is performed according to the key frames and the reference satellite image data, and single-frame absolute positioning is performed according to the heterogeneous image matching results to obtain the single-frame absolute positioning results, and inter-frame relative positioning is performed according to the single-frame absolute positioning results and non-key frames to obtain the position of the unmanned aerial vehicle. The whole process is short in time and high in precision, and a heterogeneous image matching algorithm with rotation invariance is adopted to ensure the matching accuracy and robustness of images with large rotation angles, and ensure the accuracy of positioning when illumination changes, perspective changes and resolution differences occur. This solves the problem that related technologies are difficult to meet both high precision and fast processing speed at the same time, that matching accuracy and robustness will drop significantly when processing images with large rotation angles, and that changes in lighting, viewing angle, and resolution have a significant impact on positioning accuracy.

[0041] Specifically, Figure 1 A flowchart of a method for rapid and accurate positioning of a UAV based on satellite image matching provided in an embodiment of the present application.

[0042] like Figure 1 As shown, the method for rapid and accurate positioning of a UAV based on satellite image matching includes the following steps:

[0043] In step S101, drone route planning information, drone data stream and reference satellite image data are obtained.

[0044] Among them, the drone route planning information includes but is not limited to key attitude information such as initial position, flight altitude, and real-time position.

[0045] It is understandable that the embodiments of the present application need to obtain drone route planning information, drone data stream and reference satellite image data as data reserves.

[0046] In step S102, key frames and non-key frames in the drone data stream are retrieved according to the drone route planning information, wherein the key frames are frames determined by the target posture information in the drone route planning information, and the non-key frames are frames other than the key frames in the drone data stream.

[0047] Among them, the target attitude is when the attitude of the drone changes significantly, such as the flight attitude when changing the heading. The key frames will be described in detail below and will not be repeated here.

[0048] It can be understood that the embodiments of the present application can retrieve the drone data stream frame information based on the drone route planning information, and retrieve the key frames determined by the target posture information in the drone route planning information, as well as the non-key frames other than the key frames in the drone data stream.

[0049] In an embodiment of the present application, the key frame includes at least one of an image taken when the drone's posture changes, an image with target texture information in the corresponding area of ​​the reference satellite image data, an image taken when the drone takes off, and an image taken at a preset interval.

[0050] It can be understood that the drone of the embodiment of the present application will capture key frames when it reaches each key point according to the route planning. The key frames include images captured when the drone's posture changes, images with target texture information in the corresponding area of ​​the reference satellite image data, images captured when the drone takes off, and images captured at preset intervals.

[0051] In step S103, heterogeneous image matching is performed based on the key frame and the reference satellite image data, single-frame absolute positioning is performed based on the heterogeneous image matching result to obtain the single-frame absolute positioning result, and inter-frame relative positioning is performed based on the single-frame absolute positioning result and the non-key frame to obtain the drone position.

[0052] Among them, the key frame and the reference satellite image are heterogeneous images, and the matching of the two is heterogeneous image matching. The process of heterogeneous image matching will be described in detail below and will not be repeated here. Single-frame absolute positioning is performed according to the heterogeneous image matching results. The process is to finely match the key frame with the reference satellite image data, and after converting the obtained points with the same name into control points, the position and attitude of the UAV are solved by using the idea of ​​collinearity of the three points of photogrammetry, namely, the center of photography, the image point, and the object point, so as to finally achieve precise positioning of the UAV.

[0053] It can be understood that the embodiments of the present application can perform heterogeneous image matching based on key frames and reference satellite image data, perform single-frame absolute positioning based on the heterogeneous image matching results, perform fine matching of the key frames with the reference satellite image data, convert the obtained points with the same name into control points, and use the idea of ​​photogrammetry that the three points of the photographic center, image points, and object points are colinear to solve the position and posture of the drone, and obtain a single-frame absolute positioning result. Based on the single-frame absolute positioning result and non-key frames, relative positioning between frames is performed to obtain the drone position.

[0054] In an embodiment of the present application, heterogeneous image matching is performed based on key frames and reference satellite image data, including: identifying a base map in the reference satellite image data; extracting first multi-level features of the base map and second multi-level features of the key frame respectively; performing feature transformation on the first multi-level features based on an intra-domain attention mechanism to obtain a first feature map, and performing feature transformation on the second multi-level features based on an inter-domain attention mechanism to obtain a second feature map; performing coarse-level matching on the first feature map and the second feature map to obtain a coarse-level matching result, and performing fine-level matching on the coarse-level matching result to obtain a fine-level matching result.

[0055] Among them, the first multi-level feature of the base image is the coarse-level feature at 1 / 8 of the original image size; the second multi-level feature of the key frame is the fine-level feature at 1 / 2 of the original image dimension; the coarse-level matching and the fine-level matching will be described in detail below and will not be repeated here; the in-domain attention calculation method is:

[0056]

[0057] The calculation method of inter-domain attention is:

[0058]

[0059] It can be understood that the embodiment of the present application can identify the base map in the reference satellite image data, extract the coarse-level features at 1 / 8 of the original image size as the first multi-level features, and extract the fine-level features at 1 / 2 of the original image dimension as the second multi-level features. Based on the above-mentioned intra-domain attention mechanism formula and inter-domain attention mechanism formula, the intra-domain attention mechanism is used to perform feature transformation on the first multi-level features to obtain a first feature map, and the inter-domain attention mechanism is used to perform feature transformation on the second multi-level features to obtain a second feature map. Finally, the first feature map and the second feature map are coarsely matched to obtain a coarse matching result, and the coarse matching result is finely matched to obtain a fine matching result.

[0060] In the embodiment of the present application, the prediction formula for coarse-level matching is:

[0061]

[0062] in, is a rough matching; MNN(﹒) is a mutual nearest neighbor criterion algorithm; P c is the confidence matrix; θ C is the confidence threshold.

[0063] Among them, the MNN mutual nearest neighbor criterion algorithm is a commonly used matching method, which can filter possible outlier rough matches to reduce the possibility of false matching; the confidence threshold is specifically set according to actual needs and is not specifically limited here.

[0064] It is understandable that the embodiment of the present application can obtain the result of coarse matching through the above formula, through coarse matching point pairs, mutual nearest neighbor criterion algorithm, confidence matrix and confidence threshold calculation.

[0065] In the embodiment of the present application, performing fine-level matching on the coarse-level matching result to obtain the fine-level matching result includes: locating the coarse-level matching result In fine-level feature maps and The position of the local window is cropped, and the cropped features in each window are transformed multiple times to generate two transformations. and The local feature map centered on and Will The center vector of All vectors in are associated to obtain a heat map, where the heat map represents Each pixel in the neighborhood of The matching probability is obtained by calculating the expectation on the probability distribution. B The final matching position with sub-pixel accuracy on Collect all matches Generate the final fine-level match M f .

[0066] It is understandable that in the embodiment of the present application, the coarse-level matching result is subjected to fine-level matching to obtain the fine-level matching result. First, the coarse-level matching result is located. In fine-level feature maps and The position is then cropped in the local window, and the cropped features in each window are transformed multiple times. The transformation can be performed using a smaller feature enhancement module. After generating two transformations, and The local feature map centered on and Will The center vector of All vectors in the expression are associated Each pixel in the neighborhood of The heat map of the matching probability is finally obtained by calculating the expectation on the probability distribution. B The final matching position with sub-pixel accuracy on Collect all matches Generate the final fine-level match M f .

[0067] In the embodiment of the present application, single-frame absolute positioning is performed according to the heterogeneous image matching result to obtain the single-frame absolute positioning result, including: obtaining the same-name control points according to the object coordinate transformation and image coordinate transformation of the heterogeneous image matching result; calculating the position of the drone according to the same-name control points, wherein the calculation formula of the position of the drone is:

[0068]

[0069] Among them, (x0, y0) is the image principal point, (x, y) is the image point coordinate, f is the principal distance, (X A ,Y A ,Z A ) is the object coordinate, (X S ,Y S ,Z S ) are the three-dimensional coordinates of the camera imaging center in the object coordinate system; a1a2······c3 are obtained from the posture information; the calculation of object coordinate transformation and image coordinate transformation will be described in detail below and will not be repeated here.

[0070] It can be understood that the embodiments of the present application can calculate the position of the UAV through the above formula through information such as the principal point, principal distance, image point coordinates, object coordinates and the three-dimensional coordinates of the camera imaging center in the object coordinate system, and obtain the control points with the same name based on the object coordinate transformation and image coordinate transformation of the heterogeneous image matching results, and finally calculate the position of the UAV based on the control points with the same name.

[0071] In the embodiment of the present application, the formula for object coordinate transformation is:

[0072] X=A·l+B·r+C

[0073] Y=D·l+E·r+F

[0074] Among them, A is the horizontal component of the pixel size in the X direction; B is the vertical component of the pixel size in the X direction (usually 0); C is the X coordinate of the upper left corner of the image; D is the vertical component of the pixel size in the Y direction (usually 0); E is the horizontal component of the pixel size in the Y direction; F is the Y coordinate of the upper left corner of the image; (l, r) is the pixel coordinate; (X, Y) is the geographic coordinate;

[0075] The formula for image coordinate transformation is:

[0076]

[0077] Among them, dx, dy represents the physical size of a pixel; (u0, v0) represents the position of the image principal point in the image plane coordinate system; (u, v) is the drone image coordinate; (x, y) is the image plane coordinate.

[0078] It can be understood that the embodiments of the present application can convert pixel coordinates into geographic coordinates through the above formula to achieve object coordinate transformation; and convert drone image coordinates into image plane coordinates to achieve image coordinate transformation.

[0079] In the embodiment of the present application, the matching method used for relative positioning between frames includes FAST corner point extraction, BRIFE descriptor calculation, Hamming distance and verification of epipolar constraints.

[0080] Among them, FAST corner extraction is a fast and effective corner detection algorithm that can extract feature points in pixels; BRIFE is a binary descriptor used to describe feature points. Its description vector consists of many 0s and 1s, encoding the brightness relationship between two random pixels near the key point; Hamming distance is a measure of the difference between two strings of equal length. For two binary descriptors, the Hamming distance is the number of different bits in the two descriptors.

[0081] It can be understood that the main steps of the inter-frame relative positioning matching method of the embodiment of the present application include FAST corner point extraction, BRIFE descriptor calculation, feature matching based on Hamming distance, using FAST corner points to extract feature points in pixels, and performing BRIFE descriptor calculation, and finally performing feature matching based on Hamming distance to verify the epipolar constraint.

[0082] According to the method for rapid and precise positioning of a UAV based on satellite image matching proposed in the embodiment of the present application, by obtaining the UAV route planning information, the UAV data stream and the reference satellite image data, the key frames and non-key frames in the UAV data stream are retrieved, heterogeneous image matching is performed according to the key frames and the reference satellite image data, and single-frame absolute positioning is performed according to the heterogeneous image matching result to obtain the single-frame absolute positioning result, and inter-frame relative positioning is performed according to the single-frame absolute positioning result and the non-key frames to obtain the UAV position. The whole process is time-saving and highly accurate, and a heterogeneous image matching algorithm with rotation invariance is adopted to ensure the matching accuracy and robustness of images with large rotation angles, and to ensure the accuracy of positioning when the illumination changes, the viewing angle changes and the resolution difference.

[0083] The following is a further description of an embodiment of rapid and accurate positioning of drones based on satellite image matching through a specific embodiment. This embodiment provides an embodiment of remote sensing image interest point matching based on twin encoding spatial attention mechanism. The overall architecture is as follows: Figure 2 As shown, the overall algorithm structure is as follows Figure 3 As shown, the following steps are included:

[0084] 1. Data Preparation

[0085] Before positioning, it is necessary to prepare the UAV route planning information (including but not limited to the initial position, flight altitude, real-time position and other key attitude information) and high-resolution base map data of the flight area, and perform data diversion by framing the real-time image data stream shot by the UAV and performing key frame detection.

[0086] 2. Single frame absolute positioning

[0087] This embodiment is applicable to the positioning of key frames in the process of drone shooting. Key frames include but are not limited to: images taken when the drone's attitude changes significantly (for example, when changing the heading), images with rich texture information in the corresponding area of ​​the base map, the first image taken when the drone takes off, and images taken at artificially specified intervals (such as 5s intervals). The main process of this embodiment is: first, the key frames taken by the drone are finely matched with the base map, and the obtained points of the same name are converted into control points. The position and attitude of the aircraft are solved by using the idea of ​​collinearity of the three points of the photography center, image point, and object point in photogrammetry, and finally the precise positioning of the drone is achieved. In addition, it should be noted that the position information (X, Y, Z) solved by this embodiment can not only be used for positioning at the current moment, but also can be used as the initial value required by the embodiment based on inter-frame matching positioning, mainly used to eliminate the cumulative error of inter-frame matching relative positioning.

[0088] 2.1 Heterogeneous Image Matching

[0089] Considering that when matching drone images with base maps, the two are heterogeneous images, there are large spatial domain differences due to changes in perspective, including: 1) Image style differences. Due to changes in factors such as lighting, shooting time, and camera shooting parameters, the image styles are very different; 2) Direction changes. The shooting directions of each image of ground, drone, and satellite views are different, and the data set does not have direction information annotations; 3) Position offset and scale changes. Generally, the resolution of drone images is much higher than that of satellite images, and the inconsistency between image scales must be overcome. In addition, the positions of target objects in the image are different, which may cause large offsets. Therefore, in order to overcome the above problems, when matching drone images with base maps, a matching implementation based on the attention mechanism in Transformer is selected. The model can be divided into four modules: feature extraction, feature conversion, coarse-level matching, and fine-level matching.

[0090] (1) Feature extraction

[0091] In this solution, a standard convolutional architecture with FPN is used to extract multi-level features from two images. and represents the coarse-level features at 1 / 8 of the original image size, and Fine-level features are at 1 / 2 the original image dimension.

[0092] Convolutional neural networks (CNNs) have translation equivariance and local inductive bias, which are very suitable for local feature extraction. The downsampling introduced by CN also reduces the input length of subsequent modules, greatly reducing the computational cost.

[0093] (2) Feature Enhancement

[0094] After local feature extraction, and The feature enhancement module extracts local features related to position and context and transforms them into features that are easy to match. and Feature enhancement adopts the method of parallel calculation of intra-domain attention and inter-domain attention in the attention module similarity calculation, and the feature map is N c After the attention calculation, a high-scale feature matching result is output.

[0095] The calculation method of in-domain attention is:

[0096]

[0097] The calculation method of inter-domain attention is:

[0098]

[0099] (3) Coarse-level matching

[0100] exist and On the image, the main direction is calculated for each feature point. The main direction can be determined by analyzing the gradient direction distribution of the pixels around the feature point. The direction corresponding to the peak of the histogram is called the main direction. The feature vector of each feature point is rotated and normalized according to its main direction so that the directions of all feature vectors are consistent. On the normalized feature map, the cosine similarity of each feature vector is calculated for rough matching to find potential matching areas. Cosine similarity is not affected by the length of the vector and only focuses on the direction, so it can better handle the scale problem when matching heterogeneous images.

[0101] Based on the confidence matrix P c Select the confidence level higher than the threshold θ C The matching is further performed using the mutual nearest neighbor criterion (MNN) to filter possible outlier rough matches, and the rough match prediction is expressed as:

[0102]

[0103] (4) Fine-level matching

[0104] After establishing the coarse matches, the coarse-to-fine module is used to refine these matches to the original image resolution. This scheme uses a correlation-based implementation. For each coarse match We first locate it in the fine-level feature map and Position on Then two groups of local windows of size w×w are cut. A smaller feature enhancement module is used to transform the features cut in each window into N f After two conversions, and The local feature map centered on and Then The center vector of All vectors in are associated to produce a heat map, which represents Each pixel in the neighborhood of By calculating the expectation on the probability distribution, we can get the matching probability of B The final matching position with sub-pixel accuracy on Collect all matches Generate the final fine-level matching N f .

[0105] 2.2 Get control points

[0106] (1) Object coordinate transformation

[0107] After matching the base map and the drone image, the information of the same-name points is generated, including the reference base map image coordinates (l, r) (unit: pixel) and the drone image coordinates (u, v) (unit: pixel, also known as frame coordinates or pixel coordinates). The reference base map is a geographic information image in Tiff format. By obtaining the geographic transformation matrix information of the image and supplementing it with DSM data, the object space coordinates (X, Y, Z) corresponding to the same-name points in the base map (l, r) can be obtained.

[0108] Given a pixel coordinate (l,r), you can use the geographic transformation matrix to convert it to geographic coordinates (X,Y). The conversion formula is as follows:

[0109] X=A·l+B·r+C

[0110] Y=D·l+E·r+F

[0111] ——Formula 6

[0112] in:

[0113] A: The horizontal component of the pixel size in the X direction.

[0114] B: The vertical component of the pixel size in the X direction (usually 0).

[0115] C: X coordinate of the upper left corner of the image.

[0116] D: The vertical component of the pixel size in the Y direction (usually 0).

[0117] E: The horizontal component of the pixel size in the Y direction (usually -1*pixel size).

[0118] F: Y coordinate of the upper left corner of the image

[0119] (2) Image coordinate transformation

[0120] Since the coordinate systems involved in the calculation of the collinear equation are the image plane coordinate system and the image space coordinate system, it is necessary to convert the drone image coordinates (u, v) of the same-name points into image plane coordinates (x, y) through the camera intrinsic parameter matrix. The conversion formula is:

[0121]

[0122] Among them, dx, dy represent the physical size of a pixel in mm, and (u0, v0) represents the position of the image principal point in the image plane coordinate system.

[0123] Assume that the coordinates of the same-name point P are: drone image coordinates (s, l) and reference base map image coordinates (x, y). Based on this correspondence, the coordinates of the same-name point (l, r, u, v) can be converted into control point coordinates (X, Y, Z, x, y).

[0124] 2.3 Calculate the drone position

[0125] The camera center of the drone, a certain image point on the drone image and its corresponding object point in the real world are on a straight line. The intersection of multiple straight lines is the camera center of the drone, which is taken as the position of the drone. This relationship can be expressed as:

[0126]

[0127] (x0, y0) is the image principal point, (x, y) is the image point coordinate, f is the principal distance, (X A ,Y A ,Z A ) is the object coordinate, (X S ,Y S ,Z S ) is the three-dimensional coordinate of the camera imaging center in the object coordinate system, m is the scale of the drone image, and a1a2······c3 is obtained from the attitude information ω,κ, can be solved by inversely through this equation (X S ,Y S ,ZS ) as the location of the drone.

[0128] In the solution process, X S , Y S , Z S , ω, κ, f, x0, y0 are unknown parameters, which can be expressed by their approximate values ​​(the same symbols are used in the derivation process) plus the corresponding correction factor ΔX S , ΔY S , ΔZ S , Δω, Δk, Δf, Δx0, Δy0 are expressed, and the general form of the linearization error equation is obtained:

[0129]

[0130] 3. Relative positioning between frames

[0131] The relative positioning between frames adopts the strategy of first fast matching between frames and then positioning. In the two matching frames, the previous frame already has accurate posture information, which can be corrected first and then matched again. The control point of the next frame is obtained through the embodiment described in 2.2, and then the position of the drone is calculated using the embodiment of 2.3 to achieve real-time positioning of the drone.

[0132] Among them, fast inter-frame matching refers to the fast matching of two frames of images taken continuously during the flight of the drone. The two images have high similarity, that is, similar lighting conditions, the same resolution, and a relatively uniform shooting angle. Therefore, in order to achieve the purpose of fast matching, the traditional matching implementation example can be selected in this solution. The main steps include FAST corner point extraction, BRIFE descriptor calculation, and feature matching based on Hamming distance.

[0133] 3.1 FAST corner extraction

[0134] Take the pixel to be matched p as the center and set its brightness to I p , select 16 pixels on a circle with a radius of 3, and set the threshold T = I p 20%, if there are N consecutive points on the selected circle with a brightness greater than I p +T or less than I p -T, then pixel p can be considered as a feature point.

[0135] 3.2 BRIFE descriptor calculation

[0136] BRIEF (Binary Robust Independent Elementary Features) is a binary descriptor whose description vector consists of many 0s and 1s that encode the brightness of two random pixels near the key point (for example, I a and I b ) size relationship: If I a Than I b If the value is large, then 1 is taken, otherwise 0 is taken. If 128 such I a , I b , then we get a 128-dimensional vector consisting of 0 and 1. The calculation process is:

[0137] 1) Take the feature point as the center and take an area A of size S x S, where the four sides of the area are perpendicular or parallel to the coordinate axis;

[0138] 2) Use a Gaussian filter with a variance of σ and a convolution kernel size of N x N to perform Gaussian smoothing on each point in area A;

[0139] 3) Build the descriptor:

[0140]

[0141] Among them, p(x) is the pixel intensity at x(u,v) after Gaussian smoothing, p(y) is the pixel intensity at y(u,v) after Gaussian smoothing, and the x,y point pair is randomly selected.

[0142] 3.3 Feature matching based on Hamming distance

[0143] Hamming distance is a measure of the difference between two strings of equal length. For two binary descriptors, the Hamming distance is the number of different bits in the two descriptors;

[0144] d H (m,n)=∑(m[i]≠n[i])——Formula 11

[0145] This distance metric is very suitable for comparing binary feature descriptors. The specific method is to compare two binary descriptors bit by bit and count the number of different bits. The smaller the Hamming distance, the more similar the two descriptors are and the higher the possibility of feature point matching.

[0146] 3.4 Verifying Epipolar Constraints

[0147] Epipolar constraint verification is used to ensure that feature point matching between two views is correct. The epipolar constraint states that if two feature points are projections of the same world point in two different views, then the two feature points must lie on the epipolar line in the other image. Verifying the epipolar constraint can significantly improve the robustness and accuracy of the system, helping to reduce the number of false matches, thereby improving the performance of the entire system.

[0148] Assume that we want to find the motion between two frames of images I1 and I2. Let the motion from the first frame to the second frame be R,t, and the centers of the two cameras be O1 and Q2 respectively. Consider that there is a feature point P1 in I1, which corresponds to feature point P2 in I2, such as Figure 4 As shown, this result is obtained by the previous feature matching. Under the premise of correct matching, P1 and P2 are the projections of the same spatial point P on two imaging planes.

[0149] The geometric relationships between them are:

[0150] 1) Connection and connection In three-dimensional space, they must intersect at P;

[0151] 2) The three points O1, O2, and P can determine a plane (polar plane);

[0152] 3) The corner points of the line O1O2 (baseline) and the image planes I1 and I2 are e1 and e2 (poles) respectively;

[0153] 4) Epipolar plane and image plane I1, I2 compared to lines l1, l2 (episodic lines).

[0154] The epipolar constraint is expressed as: p2 T K -T t ∧ R -1 p1=0 (not deduced here). Its geometric meaning is that O1, O2, and P are coplanar. K is the intrinsic matrix of the camera, R is the rotation matrix of the camera, and t is the translation matrix of the camera. The middle part is recorded as two matrices: Basic matrix: E=t ∧ R, essential matrix F = K -T EK -1 , so the epipolar constraint is simplified to

[0155] E=t∧R,F=K -T EK -1 , x2 T Ex1=p2 T Fp1=0——Formula 11

[0156] Among them, x1 and x2 are the coordinates of two pixels on the normalized plane.

[0157] like Figure 5 The results of other algorithms (LoFTR) of this embodiment are shown as follows: Figure 6 The results of the heterogeneous image matching algorithm with rotation invariance in this embodiment are shown. It can be seen from the above two result figures that when LoFTR is directly used for heterogeneous image matching tasks, the matched points are sparse, unevenly distributed, and there are many mismatched points. It is not robust in weak texture areas such as forests, which will have a great impact on the final positioning accuracy; the heterogeneous image matching algorithm with rotation invariance in this embodiment can almost accurately match each pair of points with the same name, and the points generated are dense and evenly distributed. It is robust even in weak texture areas such as forests, and can be well applied to positioning tasks, ensuring high-precision positioning requirements.

[0158] Next, a device for rapid and precise positioning of a UAV based on satellite image matching proposed in accordance with an embodiment of the present application will be described with reference to the accompanying drawings.

[0159] Figure 7 It is a block diagram of a device for rapid and accurate positioning of a UAV based on satellite image matching according to an embodiment of the present application.

[0160] like Figure 7 As shown, the device 10 for quickly and accurately positioning a UAV based on satellite image matching includes: an acquisition module 201 , a retrieval module 202 and an execution module 203 .

[0161] Among them, the acquisition module 201 is used to obtain the UAV route planning information, the UAV data stream and the reference satellite image data; the retrieval module 202 is used to retrieve the key frames and non-key frames in the UAV data stream according to the UAV route planning information, wherein the key frames are frames determined by the target posture information in the UAV route planning information, and the non-key frames are frames other than the key frames in the UAV data stream; the execution module 203 is used to perform heterogeneous image matching according to the key frames and the reference satellite image data, perform single-frame absolute positioning according to the heterogeneous image matching results to obtain the single-frame absolute positioning results, and perform relative positioning between frames according to the single-frame absolute positioning results and the non-key frames to obtain the UAV position.

[0162] In an embodiment of the present application, the key frame includes at least one of an image taken when the drone's posture changes, an image with target texture information in the corresponding area of ​​the reference satellite image data, an image taken when the drone takes off, and an image taken at a preset interval.

[0163] In an embodiment of the present application, the execution module 203 is further used to: perform heterogeneous image matching according to the key frame and the reference satellite image data, including: identifying the base map in the reference satellite image data; extracting the first multi-level features of the base map and the second multi-level features of the key frame respectively; performing feature transformation on the first multi-level features based on the intra-domain attention mechanism to obtain a first feature map, and performing feature transformation on the second multi-level features based on the inter-domain attention mechanism to obtain a second feature map; performing coarse-level matching on the first feature map and the second feature map to obtain a coarse-level matching result, and performing fine-level matching on the coarse-level matching result to obtain a fine-level matching result.

[0164] In the embodiment of the present application, the prediction formula for coarse-level matching is:

[0165]

[0166] in, is a rough matching; MNN(﹒) is a mutual nearest neighbor criterion algorithm; P c is the confidence matrix; θ C is the confidence threshold.

[0167] In the embodiment of the present application, performing fine-level matching on the coarse-level matching result to obtain the fine-level matching result includes: locating the coarse-level matching result In fine-level feature maps and The position of the local window is cropped, and the cropped features in each window are transformed multiple times to generate two transformations. and The local feature map centered on and Will The center vector of All vectors in are associated to obtain a heat map, where the heat map represents Each pixel in the neighborhood of The matching probability is obtained by calculating the expectation on the probability distribution. B The final matching position with sub-pixel accuracy on Collect all matches Generate the final fine-level match M f .

[0168] In the embodiment of the present application, the execution module 203 is further used to: perform single-frame absolute positioning according to the heterogeneous image matching result to obtain a single-frame absolute positioning result, including: obtaining the same-name control points according to the object coordinate transformation and image coordinate transformation of the heterogeneous image matching result; and calculating the position of the drone according to the same-name control points, wherein the calculation formula of the position of the drone is:

[0169]

[0170] Among them, (x0, y0) is the image principal point, (x, y) is the image point coordinate, f is the principal distance, (X A ,Y A ,Z A ) is the object coordinate, (X S ,Y S ,Z S ) is the three-dimensional coordinate of the camera imaging center in the object coordinate system; a1a2······c3 are obtained from the posture information.

[0171] In the embodiment of the present application, the formula for object coordinate transformation is:

[0172] X=A·l+B·r+C

[0173] Y=D·l+E·r+F

[0174] Among them, A is the horizontal component of the pixel size in the X direction; B is the vertical component of the pixel size in the X direction (usually 0); C is the X coordinate of the upper left corner of the image; D is the vertical component of the pixel size in the Y direction (usually 0); E is the horizontal component of the pixel size in the Y direction; F is the Y coordinate of the upper left corner of the image; (l, r) is the pixel coordinate; (X, Y) is the geographic coordinate; the formula for image coordinate transformation is:

[0175]

[0176] Among them, dx, dy represents the physical size of a pixel; (u0, v0) represents the position of the image principal point in the image plane coordinate system; (u, v) is the drone image coordinate; (x, y) is the image plane coordinate.

[0177] In the embodiment of the present application, the matching method used for relative positioning between frames includes FAST corner point extraction, BRIFE descriptor calculation, Hamming distance and verification of epipolar constraints.

[0178] It should be noted that the aforementioned explanation of the embodiment of the method for rapid and precise positioning of a drone based on satellite image matching is also applicable to the device for rapid and precise positioning of a drone based on satellite image matching in this embodiment, and will not be repeated here.

[0179] According to the device for rapid and precise positioning of unmanned aerial vehicles based on satellite image matching proposed in the embodiment of the present application, through the coordinated action of the acquisition module, the retrieval module and the execution module, it is possible to obtain the unmanned aerial vehicle route planning information, the unmanned aerial vehicle data stream and the reference satellite image data, retrieve the key frames and non-key frames in the unmanned aerial vehicle data stream, perform heterogeneous image matching according to the key frames and the reference satellite image data, perform single-frame absolute positioning according to the heterogeneous image matching results to obtain the single-frame absolute positioning results, perform inter-frame relative positioning according to the single-frame absolute positioning results and the non-key frames to obtain the position of the unmanned aerial vehicle. The entire process is time-saving and highly accurate, and adopts a heterogeneous image matching algorithm with rotation invariance to ensure the matching accuracy and robustness of images with large rotation angles, and ensure the accuracy of positioning when the illumination changes, the viewing angle changes and the resolution difference.

[0180] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0181] A memory 301 , a processor 302 , and a computer program stored in the memory 301 and executable on the processor 302 .

[0182] When the processor 302 executes the program, the method for quickly and accurately positioning a UAV based on satellite image matching provided in the above embodiment is implemented.

[0183] Furthermore, the electronic device also includes:

[0184] The communication interface 303 is used for communication between the memory 301 and the processor 302 .

[0185] The memory 301 is used to store computer programs that can be run on the processor 302 .

[0186] The memory 301 may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.

[0187] If the memory 301, the processor 302 and the communication interface 303 are implemented independently, the communication interface 303, the memory 301 and the processor 302 can be connected to each other through a bus and communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0188] Optionally, in a specific implementation, if the memory 301, the processor 302 and the communication interface 303 are integrated on a chip, the memory 301, the processor 302 and the communication interface 303 can communicate with each other through an internal interface.

[0189] The processor 302 may be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

[0190] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0191] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0192] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0193] It should be understood that the various parts of the present application can be implemented in hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, the steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array, a field programmable gate array, etc.

[0194] A person of ordinary skill in the art may understand that all or part of the steps carried by the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the above-mentioned program may be stored in a computer-readable storage medium, which, when executed, includes one of the steps of the method embodiment or a combination thereof.

[0195] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching, characterized in that: The following steps are involved: Obtain UAV route planning information, UAV data streams, and reference satellite image data; Retrieve key frames and non-key frames in the drone data stream according to the drone route planning information, wherein the key frames are frames determined by the target posture information in the drone route planning information, and the non-key frames are frames other than key frames in the drone data stream; Heterogeneous image matching is performed according to the key frame and the reference satellite image data, single-frame absolute positioning is performed according to the heterogeneous image matching result to obtain a single-frame absolute positioning result, and inter-frame relative positioning is performed according to the single-frame absolute positioning result and non-key frames to obtain the position of the drone.

2. The method for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching according to claim 1 is characterized in that: The key frame includes at least one of an image taken when the drone's posture changes, an image with target texture information in a corresponding area of ​​the reference satellite image data, an image taken when the drone takes off, and an image taken at a preset interval.

3. The method for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching according to claim 1 is characterized in that: The performing heterogeneous image matching according to the key frame and the reference satellite image data includes: Identifying a base map in the reference satellite image data; Extracting first multi-level features of the base image and second multi-level features of the key frame respectively; Performing feature transformation on the first multi-level features based on an intra-domain attention mechanism to obtain a first feature map, and performing feature transformation on the second multi-level features based on an inter-domain attention mechanism to obtain a second feature map; A coarse matching is performed on the first feature map and the second feature map to obtain a coarse matching result, and a fine matching is performed on the coarse matching result to obtain a fine matching result.

4. The method for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching according to claim 3 is characterized in that: The prediction formula for the coarse-level matching is: in, is the rough matching result; MNN(﹒) is the mutual nearest neighbor criterion algorithm; P c is the confidence matrix; θ C is the confidence threshold.

5. The method for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching according to claim 3 is characterized in that: The performing fine-level matching on the coarse-level matching result to obtain a fine-level matching result includes: Locating coarse-level matching results In fine-level feature maps and location; Crop the local window and transform the features cropped in each window multiple times, generating two transformations respectively. and The local feature map centered on and Will The center vector of All vectors in are associated to obtain a heat map, where the heat map represents Each pixel in the neighborhood of The matching probability of By calculating the expectation on the probability distribution, we can get B The final matching position with sub-pixel accuracy on Collect all matches Generate the final fine-level match M f .

6. The method for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching according to claim 1, characterized in that: The performing of single-frame absolute positioning according to the heterogeneous image matching result to obtain the single-frame absolute positioning result includes: Obtaining control points with the same name according to the object coordinate transformation and image coordinate transformation of the heterogeneous image matching result; The position of the drone is calculated according to the control points with the same name, wherein the calculation formula of the position of the drone is: Among them, (x0, y0) is the image principal point, (x, y) is the image point coordinate, f is the principal distance, (X A ,Y A ,Z A ) is the object coordinate, (X S ,Y S ,Z S ) is the three-dimensional coordinate of the camera imaging center in the object coordinate system; a1a2······x3 is obtained from the posture information.

7. The method for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching according to claim 6 is characterized in that: The formula for object coordinate transformation is: X=A·l+B·r+C Y=D·l+E·r+F Among them, A is the horizontal component of the pixel size in the X direction; B is the vertical component of the pixel size in the X direction (usually 0); C is the X coordinate of the upper left corner of the image; D is the vertical component of the pixel size in the Y direction (usually 0); E is the horizontal component of the pixel size in the Y direction; F is the Y coordinate of the upper left corner of the image; (l, r) is the pixel coordinate; (X, Y) is the geographic coordinate; The formula for the image coordinate transformation is: Among them, dx, dy represents the physical size of a pixel; (u0, v0) represents the position of the image principal point in the image plane coordinate system; (u, v) is the drone image coordinate; (x, y) is the image plane coordinate.

8. The method for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching according to claim 1, characterized in that: The matching method used in the relative positioning between frames includes FAST corner point extraction, BRIFE descriptor calculation, Hamming distance and verification of epipolar constraints.

9. A device for rapid and accurate positioning of unmanned aerial vehicles based on satellite image matching, characterized in that: include: The acquisition module is used to obtain the UAV route planning information, UAV data stream and reference satellite image data; A retrieval module, used to retrieve key frames and non-key frames in the drone data stream according to the drone route planning information, wherein the key frames are frames determined by the target posture information in the drone route planning information, and the non-key frames are frames other than the key frames in the drone data stream; The execution module is used to perform heterogeneous image matching according to the key frame and the reference satellite image data, perform single-frame absolute positioning according to the heterogeneous image matching result to obtain a single-frame absolute positioning result, and perform inter-frame relative positioning according to the single-frame absolute positioning result and non-key frames to obtain the position of the drone.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for rapid and precise positioning of a drone based on satellite image matching as described in any one of claims 1 to 8.

Images