GPS-based aerial image rapid pose estimation method and device, and medium

Abstract

The invention provides a GPS-based aerial image rapid pose estimation method and device and a medium, and the method comprises the steps: carrying out the rapid screening from a disordered image set based on the GPS position information of an image, and constructing a stereo image pair meeting a preset geometric constraint; carrying out downsampling on original images in the stereo image pair to generate small images with low resolution, and carrying out feature matching on the small images to obtain connection point information and estimate a homography relation between the images; calculating an image space rotation angle according to the homography relation, and performing combined calculation by combining a ground coordinate system azimuth angle calculated by GPS position data so as to obtain a rotation deflection angle estimated value of the image; and calculating complete exterior orientation elements of the image by taking the rotation deflection angle estimated value and the GPS position as initial values. According to the method, high-efficiency pose estimation independent of expensive hardware is realized, the calculation redundancy is greatly reduced, the real-time requirements of scenes such as rapid mapping and emergency response are met, and the convergence speed and stability of a traditional precision processing flow are supported and improved.

Description

Technical Field

[0001] This invention belongs to the field of photogrammetry technology, and in particular relates to a technical solution for rapidly estimating the pose of aerial images using Global Positioning System (GPS) information. Background Technology

[0002] In the fields of aerial photogrammetry and computer vision, accurate and rapid estimation of image pose has always been a core technical aspect for achieving high-precision 3D reconstruction and geographic information extraction. Traditional pose estimation methods mainly rely on the Structure from Motion (SfM) framework. This framework generates a massive number of connection points by comprehensively extracting and matching feature points across a large number of images, and then uses the bundle adjustment algorithm to solve for the interior and exterior orientation elements of all images. Although this method can achieve high accuracy under ideal conditions, its inherent shortcomings become increasingly apparent when dealing with the large-scale, disordered, and even real-time image data generated by modern aerial remote sensing: First, the matching strategy of all connection points leads to a combinatorial increase in computational complexity, generating far more connection points than necessary, resulting in huge computational redundancy and time consumption; second, bundle adjustment heavily relies on good initial values. For large-scale disordered datasets, the initialization process is complex and unstable, and incorrect matching can easily lead to divergence or getting trapped in local optima; most importantly, the processing efficiency of traditional methods is difficult to meet the needs of application scenarios with extremely high timeliness requirements, such as emergency mapping and real-time monitoring.

[0003] To address the aforementioned issues, existing technologies attempt to incorporate GPS / IMU integrated navigation systems for assisted positioning and attitude determination. While high-precision IMUs can directly provide attitude information, their hardware costs are extremely high, and the post-processing procedures, such as spatiotemporal synchronization of the IMU with image data and error compensation, are exceptionally complex, significantly limiting the widespread adoption of this technology in mass-market, low-cost applications. Therefore, how to quickly and robustly estimate the attitude parameters of images, especially the kappa angle, which is most critical for subsequent processing, using only positioning information provided by ordinary GPS modules, has long been a technical bottleneck that has remained unresolved in this field.

[0004] The core of the current technological dilemma lies in the failure to break through the traditional paradigm of "global optimization - high computational load." Existing solutions either sacrifice efficiency in pursuit of global optimization or rely on expensive hardware at the expense of universality, lacking a lightweight solution capable of rapidly obtaining reliable initial pose values ​​under limited resource conditions. This technological gap severely restricts the in-depth development of emerging technologies such as UAV remote sensing and mobile mapping in real-time response and large-scale application scenarios. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a rapid pose estimation technique for aerial imagery based on GPS information.

[0006] The above-mentioned technical problems of the present invention are mainly solved by the following technical solutions: This invention provides a method for rapid pose estimation based on GPS aerial imagery, comprising: Based on GPS location information from images, stereo image pairs that meet preset geometric constraints are quickly selected from unordered image sets. The original images in the stereo image pair are downsampled to generate low-resolution small images, and feature matching is performed on the small images to obtain connection point information and estimate the homography relationship between the images; The image rotation angle is calculated based on the homography relationship, and combined with the ground coordinate system azimuth angle calculated from the GPS position data, a joint calculation is performed to obtain the image deflection angle estimate; the image complete exterior orientation elements are calculated using the deflection angle estimate and the GPS position as initial values.

[0007] Furthermore, the specific methods for quickly filtering and constructing stereo image pairs based on GPS location information from images include: All GPS coordinates of the images are uniformly converted to a local Cartesian coordinate system; Set a usage status indicator for each image; Traverse the unused images and calculate their spatial distance from other unused images in the local Cartesian coordinate system. Based on preset baseline length thresholds and image overlap thresholds, candidate image pairs are geometrically constrained for selection. Select the image that meets the constraints and is spatially closest to the current image to form a stereo image pair, and update the usage status of both images. Iterate through the above process until a set of stereo image pairs without redundancy is generated.

[0008] Moreover, the image overlap in the geometric constraint screening is quickly estimated based on the GPS distance between images, camera focal length, regional average flight altitude, and camera sensor size.

[0009] Furthermore, the specific methods for downsampling the original image to generate a low-resolution small image include: Each original image in the stereo pair is scaled down to a fixed low-resolution size using an interpolation method to generate a small image for fast matching.

[0010] Furthermore, specific methods for implementing feature matching on small, low-resolution images include: Extract feature points and descriptors from small images; A strategy based on descriptor distance ratios is used for initial feature point matching. The random sampling consensus algorithm is applied to robustly estimate the homography matrix between images from the initial matching results.

[0011] Furthermore, the calculation of the image-side rotation angle based on the homography relationship specifically includes: Using the homography matrix, the projection position of the principal point of one image in the coordinate system of another image's smaller image is calculated; The image plane rotation angle is calculated based on the offset of the projection position relative to the center of the small image.

[0012] Furthermore, after obtaining the estimated spin angle, it is used as an initial value and input into the motion recovery structure processing flow to improve the convergence speed of the bundle adjustment process.

[0013] On the other hand, the present invention also 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 a GPS-based rapid pose estimation method for aerial images as described above.

[0014] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a GPS-based rapid pose estimation method for aerial images as described above.

[0015] On the other hand, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a GPS-based rapid pose estimation method for aerial images as described above.

[0016] Through the above technical solution, this invention proposes a rapid pose estimation technique for aerial imagery based on GPS and tie points. It achieves rapid extraction and construction of stereo image pairs using GPS location information, rapidly matches tie points using downsampled small images, calculates the image plane rotation angle based on the position coordinates of the principal point in the other image, and calculates the azimuth angle in the ground coordinate system using GPS location data. The angle observations in both coordinate systems are jointly solved to obtain the estimated rotation angle of the image. Finally, using the estimated rotation angle and GPS position as initial values, a simplified resection method is used to calculate the complete exterior orientation elements. The pose parameters are verified and accuracy evaluated through coplanar condition equations and forward resection. The core innovation lies in constructing a novel technical route from GPS-guided stereo image pair extraction and rapid small image matching to joint image-object-side calculation, achieving rapid and robust estimation of image rotation angle and complete pose parameters without relying on high-cost hardware and complex global optimization.

[0017] Compared with the prior art, the technical solution of the present invention has the following breakthrough advantages: by using an innovative mechanism based on GPS to extract the fewest stereo image pairs, the computational redundancy of traditional methods is fundamentally reduced; by adopting a small image matching strategy, the computational load of feature matching is reduced; and the "joint calculation of image-side rotation angle and object-side azimuth angle" model realizes the direct and rapid estimation of the rotation angle using only ordinary GPS and image data.

[0018] It is particularly noteworthy that this invention not only serves as an independent pose estimation method to meet the demands of rapid processing, but its high-precision rotation angle estimate can also be used as a high-quality initial value for traditional precise SfM processing, significantly improving the convergence speed of bundle adjustment and shortening the overall processing time while maintaining accuracy. This characteristic makes this invention irreplaceable in applications such as rapid mapping, emergency response, and large-scale 3D reconstruction, laying a solid foundation for achieving the technological leap from "high precision" to "high efficiency + high precision". Attached Figure Description

[0019] Figure 1 This is an overall flowchart of the method of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Example 1 This invention provides a method for rapid pose estimation of aerial imagery based on GPS information. Through three core steps—GPS-guided stereo image pair extraction, rapid matching of low-resolution small images, and GPS estimation of rotation angles from tie points and images—it achieves rapid and robust estimation of image rotation angles and complete exterior orientation elements, relying solely on ordinary GPS observations.

[0022] An improved GPS-based fast pose estimation method for aerial imagery, as described in this embodiment, includes: Based on GPS location information from images, stereo image pairs that meet preset geometric constraints are quickly selected from unordered image sets. The original images in the stereo image pair are downsampled to generate low-resolution small images, and feature matching is performed on the small images to obtain connection point information and estimate the homography relationship between the images; The image rotation angle is calculated based on the homography relationship, and combined with the ground coordinate system azimuth angle calculated from the GPS position data, a joint calculation is performed to obtain the image deflection angle estimate; the image complete exterior orientation elements are calculated using the deflection angle estimate and the GPS position as initial values.

[0023] Furthermore, based on GPS location information, a set of stereo image pairs without redundancy can be quickly filtered and constructed through spatial distance and overlap constraints.

[0024] This invention achieves efficient pose estimation without relying on expensive hardware, greatly reducing computational redundancy. It can not only meet the real-time requirements of scenarios such as rapid mapping and emergency response, but its high-precision initial values ​​can also significantly improve the convergence speed and stability of traditional precision processing.

[0025] Example 2 like Figure 1 As shown in the figure, the flowchart of a GPS-based rapid pose estimation method for aerial imagery provided by this embodiment of the invention includes the following steps: Step 1: Rapid extraction and construction of stereo image pairs based on GPS location information; Step 2: A low-resolution small image map is generated by downsampling the original image to achieve rapid extraction and matching of connection points; Step 3: Combine image-side and object-side information to achieve accurate estimation of the rotation angle. In one feasible implementation, the specific steps of extracting and constructing the minimum number of stereo image pairs based on GPS location information in step 1 include: Step 1.1: First, the unordered aerial imagery undergoes GPS coordinate system unification processing. The GPS coordinates of all imagery are transformed using a local tangent plane approximate cylindrical projection to a unified local plane rectangular coordinate system (i.e., the target plane coordinate system), thus unifying the coordinate systems. The coordinate transformation formula is: (1) In the formula, ( The coordinates are in the local tangent plane coordinate system. , They are longitude and latitude, respectively. , The coordinates of the projection origin are... , The radii of curvature of the meridian and the east-west circle are... This represents the elevation value.

[0026] Step 1.2: Initialize a state tag for each image in the dataset, forming a state set S = { | i=1,2,...,n}, where Let i represent the image status, n represent the image identifier, and n represent the total number of images. Initially, all images are marked as "unused".

[0027] Step 1.3: Iterate through each image marked "unused". Calculate its comparison with all other images marked "unused". GPS spatial Euclidean distance between The distance calculation formula is: (2) In the formula, ( ) and( ( ) are images respectively and The three-dimensional coordinates in the target plane coordinate system.

[0028] Step 1.4 involves applying multiple geometric constraints, including baseline length and image overlap, to all candidate image pairs calculated in Step 1.3 to ensure that the extracted stereo image pairs have feasible photogrammetric geometric relationships. The baseline length of the stereo image pair... The constraints are: (3) In the formula, This is the minimum baseline length threshold, used to avoid insufficient forward intersection accuracy due to an excessively short baseline. The maximum baseline length threshold is determined by the average flight altitude and the image field of view, and is used to ensure that there is sufficient overlap between two images.

[0029] Image overlap constraints are rapidly estimated using a simplified model based on photogrammetric geometry. Overlap The calculation formula is as follows: (4) In the formula, The GPS spatial Euclidean distance between images is calculated in step 1.3. This refers to the camera's focal length. This represents the regional average flight altitude. This represents the physical width of the camera sensor in the direction of flight. / Approximately the width of a single image covering the ground. G The width of the overlapping region is G - The degree of overlap is ( G - ) / G 。 The overlap filtering criteria are: (5) In the formula, The preset overlap threshold is usually set to 0.6 (i.e., 60% overlap) to ensure that there is enough common area for reliable matching.

[0030] Step 1.5, for the current image From all candidate images that meet the above constraints, select the one with spatial distance. Recent images As its stereo pair. Then, the image... and The status flag is updated to "used", and the pair ( , Record it into the set of stereo image pairs.

[0031] Step 1.6: Repeat steps 1.2 to 1.5 until all images are marked as "used" or there are no unused image pairs that satisfy the constraints. The final output set of stereo image pairs S is the completed set of minimal image pairs without redundancy.

[0032] In one feasible implementation, the specific steps of the connection point matching based on the low-resolution small image in step 2 include: Step 2.1: For each stereo image pair generated in Step 1, the original image is downsampled to generate an N×N pixel mini-image (where N is between 256 and 1024; experiments show that when N ≥ 512, the number of matching points decreases by <5% compared to full resolution, while computation time is reduced by more than 80%; further increasing N provides limited accuracy improvement but significantly increases memory consumption; therefore, this embodiment preferably uses N = 512, but in practice, it can be adjusted within the range of 256–1024 according to the original image resolution and hardware conditions) to reduce the computational complexity of subsequent feature matching. Downsampling uses bilinear interpolation, and the calculation formula is: (6) In the formula, For small image images, This is the original image. These are the pixel coordinates of the small image. The pixel coordinates corresponding to the original image are calculated using the scaling factor: (7) In the formula, and These are the width and height of the original image, respectively. and These are the horizontal and vertical scaling factors, respectively. Bilinear interpolation is achieved through a weighted average of neighboring pixels, ensuring image smoothness.

[0033] Step 2.2, Feature Point Extraction. Feature points and their descriptors are extracted from the small image using a feature description algorithm. Feature point extraction is based on image gradient or intensity changes, and descriptor generation represents the local image features around the feature points. Specifically, feature point detection can be achieved using the difference of Gaussian operator or similar methods, and the descriptor is calculated as a gradient direction histogram or binary pattern within the neighborhood of the feature point. Let the feature point set be F = { | i=1,2,...,n}, where n represents the number of feature points extracted from a single downsampled image, and each feature point... Includes position coordinates ( ) and descriptor vector .

[0034] Step 2.3, Feature Matching. Feature matching is performed using the nearest neighbor ratio test. For small images... For each feature point in the image, calculate its relationship with the smaller image. The Euclidean distance of all feature point descriptors is used to find the nearest and second nearest neighbor feature points. If the ratio of the nearest neighbor distance to the second nearest neighbor distance is less than a set threshold, the matching pair is accepted as a connection point.

[0035] Step 2.4, Homography Matrix Estimation. The RANSAC algorithm is applied to robustly estimate the homography matrix H. For a matching point pair, the points... Passing in homogeneous coordinates Transform to : (7) In the formula, The RANSAC algorithm randomly selects four pairs of matching points, calculates the initial H matrix, then calculates the projection error of all matching point pairs, selects inliers with errors less than a threshold, and iteratively optimizes to obtain the optimal H matrix. The least squares method is used to solve the H matrix. Given n pairs of matching points, the equation is constructed as follows: (8) In the formula, A is a 2n×9 matrix composed of point pairs, and h is the vector form of matrix H. h is solved by singular value decomposition (SVD).

[0036] In one feasible implementation, step 3, which involves accurately estimating the deflection angle based on the connection point and GPS, specifically includes: Step 3.1, calculate the principal point position. Assume the image... The principal point of the image is in homogeneous coordinates in the coordinate system of a low-resolution small image. ,For example The homography matrix H is used to calculate its position in the image. Projection position in: (9) in, For the principal point in the image Homogeneous coordinates in the small image coordinate system, where u and v are the corresponding pixel row and column numbers, and w is the homogeneous depth factor. The projection result is normalized to obtain the image. The actual coordinates in the image plane are: (10) In the formula, That is, images The principal point in the image The corresponding positions in the image are used for subsequent connection point optimization or geometric correction. Similarly, the image can be calculated. The principal point in the image Position coordinates in .

[0037] Step 3.2, Image plane rotation angle calculation. Establish an image plane coordinate system with the center of image I as the origin, with the X-axis horizontally to the right and the Y-axis vertically upward. Based on the principal point position offset calculated in Step 3.1, calculate the image plane rotation angle k of image I relative to image J. For image J... Its image plane rotation angle The calculation formula is: Where dx = .x - dy = .y,( , () represents the coordinates of the principal point in the smaller image. These are the coordinates of the principal point's projection onto the other image. For example, The calculation is as follows: (11) Where dx is the horizontal offset of the principal point relative to the center of the small image (256, 256) (positive to the right); dy is the vertical offset of the principal point relative to the center of the small image (positive upwards).

[0038] Step 3.3, calculate the azimuth angle in the ground coordinate system. In the ground coordinate system (usually the northeast-sky coordinate system), calculate the azimuth angle based on the GPS position data of the two images. Let the coordinates of images I1 and I2 in the ground coordinate system be (…). )and( If the azimuth angle in the ground coordinate system pointing from image I1 to image I2 is... The calculation formula is: (12) Where ΔX is the difference in the eastern GPS coordinates of the two images, and ΔY is the difference in the northern GPS coordinates of the two images; and( These are the horizontal positions in the northeast-central coordinate system. This azimuth angle... It indicates the angle of rotation clockwise from due north to the baseline direction.

[0039] Step 3.4, Rotation Angle Calculation and Normalization. Based on the image plane rotation angle and ground azimuth angle calculated in Steps 3.2 and 3.3, the estimated rotation angle of the image is directly calculated: (13) Where k is the rotation angle. Considering the periodicity of angle calculation, the rotation angle is normalized to ensure that the result falls within (k=k). Within the interval [π, π].

[0040] Step 3.5, repeat steps 3.2 to 3.4 to calculate the rotation angle of image I2 pointing to image I1.

[0041] In one possible embodiment, after step 3.5, the method further includes a step of verifying and evaluating the accuracy of the solved pose parameters through coplanar condition equations and forward intersection.

[0042] In practice, the above process can be automated using computer software technology. Users only need to input the original images and corresponding GPS files, and the software can automatically output the high-precision exterior orientation elements of each image in the target coordinate system for direct use in subsequent SfM or oblique photogrammetry processing.

[0043] In aerial photogrammetry and 3D reconstruction tasks, the rationality of the initial pose parameters of an image directly affects the convergence speed and stability of subsequent bundle adjustment. The pitch and roll angles in an image are typically close to horizontal, while the yaw angle is usually related to flight attitude adjustments or camera mounting angles. The yaw angle estimation achieved in this invention provides a high-confidence initial direction value for each image, effectively solving the problem of inefficient iterations or even convergence to local optima in the adjustment process. This improves the efficiency of large-scale data processing and is suitable for modern photogrammetric applications with strict requirements for timeliness and computational cost, providing a reliable technical path for the rapid and intelligent processing of massive amounts of imagery.

[0044] Example 3 This invention provides a GPS-based rapid pose estimation system for aerial imagery, comprising the following modules: The first module is used to quickly filter and construct stereo image pairs that meet preset geometric constraints from an unordered image set based on GPS location information from images. The second module is used to downsample the original image in the stereo image pair to generate a low-resolution small image, and perform feature matching on the small image to obtain connection point information and estimate the homography relationship between the images. The third module is used to calculate the image rotation angle based on the homography relationship, and combine it with the ground coordinate system azimuth angle calculated from the GPS position data to perform joint calculation to obtain the image deflection angle estimate; using the deflection angle estimate and GPS position as initial values, the complete exterior orientation elements of the image are calculated.

[0045] In specific implementation, the method proposed in the technical solution of this invention can be automatically executed by those skilled in the art using computer software technology. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of the technical solution of this invention and computer equipment including the computer program running the corresponding computer program, should also be within the protection scope of this invention.

[0046] The following embodiments describe the electronic device provided by the present invention. The electronic device described below can be referred to in correspondence with the GPS-based aerial image fast pose estimation method described above.

[0047] The electronic device may include a processor, a communications interface, memory, and a communication bus. The processor, communications interface, and memory communicate with each other via the communication bus. The processor can call logical instructions from the memory to execute a GPS-based rapid pose estimation method for aerial imagery, primarily including the software processing steps described above.

[0048] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0049] On the other hand, embodiments of the present invention also provide a computer program product, the computer program product including a computer program, the computer program being stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute the software processing part of the GPS-based rapid pose estimation method for aerial images provided by the above methods.

[0050] In another aspect, embodiments of the present invention also provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the software processing portion of the GPS-based rapid pose estimation method for aerial images provided by the above methods.

[0051] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0052] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A GPS-based aerial image fast pose estimation method, characterized in that, include: Based on GPS location information from images, stereo image pairs that meet preset geometric constraints are quickly selected from unordered image sets. The original images in the stereo image pair are downsampled to generate low-resolution small images, and feature matching is performed on the small images to obtain connection point information and estimate the homography relationship between the images; The image rotation angle is calculated based on the homography relationship, and combined with the ground coordinate system azimuth angle calculated from the GPS position data, a joint calculation is performed to obtain the image's deflection angle estimate; using the deflection angle estimate and GPS position as initial values, the complete exterior orientation elements of the image are calculated. 2.The GPS-based aerial image rapid pose estimation method of claim 1, wherein: The specific implementation methods for quickly filtering and constructing stereo image pairs based on GPS location information from images include: All GPS coordinates of the images are uniformly converted to a local Cartesian coordinate system; Set a usage status indicator for each image; Traverse the unused images and calculate their spatial distance from other unused images in the local Cartesian coordinate system. Based on preset baseline length thresholds and image overlap thresholds, candidate image pairs are geometrically constrained for selection. Select the image that meets the constraints and is spatially closest to the current image to form a stereo image pair, and update the usage status of both images. Iterate through the above process until a set of stereo image pairs without redundancy is generated. 3.The GPS-based aerial image rapid pose estimation method of claim 2, wherein: The image overlap in the geometric constraint screening is quickly estimated based on the GPS distance between images, camera focal length, regional average flight altitude, and camera sensor size.

4. The method for fast pose estimation of aerial images based on GPS according to claim 1, characterized in that: The specific methods for downsampling the original image to generate a smaller, lower-resolution image include: Each original image in the stereo pair is scaled down to a fixed low-resolution size using an interpolation method to generate a small image for fast matching.

5. A method for rapid pose estimation of aerial imagery based on GPS according to claim 1 or 4, characterized in that: The specific methods for performing feature matching on small, low-resolution images include: Extract feature points and descriptors from small images; A strategy based on descriptor distance ratios is used for initial feature point matching. The random sampling consensus algorithm is applied to robustly estimate the homography matrix between images from the initial matching results.

6. The method for fast pose estimation of aerial images based on GPS according to claim 1, characterized in that: The calculation of the image-side rotation angle based on homography specifically includes: Using the homography matrix, calculate the projection position of the principal point of one image in the coordinate system of another image's smaller image; The image plane rotation angle is calculated based on the offset of the projection position relative to the center of the small image.

7. The method for fast pose estimation of aerial images based on GPS according to claim 1, characterized in that: After obtaining the estimated spin angle, it is used as an initial value and input into the motion recovery structure processing flow to improve the convergence speed of the bundle adjustment process.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements a GPS-based rapid pose estimation method for aerial imagery as described in any one of claims 1 to 7.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements a GPS-based rapid pose estimation method for aerial imagery as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by the processor, it implements a GPS-based rapid pose estimation method for aerial imagery as described in any one of claims 1 to 7.

Images