Drone aerial image geographical registration method, system and product

Abstract

The unmanned aerial vehicle aerial image geographic registration method, system and product belong to the technical field of remote sensing image processing and geographic information system, and solve the technical problems that the prior art depends on ground control point arrangement and post-processing manual registration, the processing procedure is complex, and the rapid response and large-scale data acquisition requirements cannot be met. Information of each unmanned aerial vehicle aerial image is read; geographic coordinates corresponding to a plurality of image feature points in each image are calculated; feature extraction and matching are performed on the images, a common view feature point set among all the images is established; a multi-view joint optimization model is constructed, the camera attitude parameters, the camera spatial position parameters and the geographic coordinates of the initial control points of each image are optimized by using the multi-view joint optimization model; a geometric transformation model is fitted, geometric correction and resampling are performed on the unmanned aerial vehicle aerial image, and the unmanned aerial vehicle aerial image after geographic registration is obtained. The method is used for realizing unmanned aerial vehicle aerial image fast automatic registration and geographic space reference giving.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing and geographic information system technology, specifically to a method, system, and product for georegistration of UAV aerial images. Background Technology

[0002] With the widespread application of UAV technology, UAV orthophotos have become an important means of acquiring high-resolution geographic information. However, the traditional UAV orthophoto acquisition process is complex, relies on the deployment of ground control points and subsequent manual registration, and has a long data processing cycle, making it difficult to meet the needs of rapid response and large-scale data acquisition.

[0003] Most drone aerial images only contain the coordinates of the center point at the time of shooting and flight parameters, lacking geospatial reference information that can be directly used for GIS (Geographic Information System) analysis, making it difficult to directly apply drone aerial images to geospatial analysis and decision-making.

[0004] In the prior art, Chinese patent document CN116363185A discloses a "georegistration method, apparatus, electronic device, and readable storage medium." This method involves stitching multiple images to obtain a stitched image, then determining the set of image points to be registered from the stitched image. Based on the pose information of each image point in the stitched image, a set of control points corresponding to the image point set is obtained. Finally, the stitched image is calibrated based on the control point set to obtain the georegistration result. However, this technical solution is mainly geared towards conventional surveying and mapping scenarios, relying on image stitching and flight strip planning. It lacks support for emergency, rapid, and low-redundancy image processing capabilities. The processing flow is complex, computationally intensive, and time-consuming, making it difficult to achieve rapid georegistration in scenarios with high timeliness requirements, such as emergency response and rapid monitoring.

[0005] In the prior art, Chinese patent document CN112132029A discloses "A rapid positioning method for UAV remote sensing images for earthquake emergency response." This method involves inputting a high spatial resolution UAV remote sensing image sequence and position and attitude parameters to obtain the exterior orientation elements for image positioning; using a world file to assist in rapid positioning of the UAV remote sensing image data, converting the exterior orientation elements of the sequence image into the position and rotation definition forms required by the world file to determine the initial image position; overlaying the initially positioned UAV remote sensing initial contour map with reference image data; and correcting the position and angle parameters using overall error improvement and rotation transformation compensation based on the overlay result to obtain the positioned sequence image. However, this technical solution relies on a high-precision POS system and planned flight path of a professional UAV, and requires prior acquisition of reference images. In sudden emergency scenarios without professional equipment, flight planning, or reference data, its technical prerequisites cannot be met. The core of this technical solution is to optimize professional processes, rather than solving the problem of rapid positioning from scratch.

[0006] Therefore, there is an urgent need for an automatic registration method based on the parameters of UAV aerial images and the image content, which can quickly convert UAV aerial images into geographic images with geospatial reference, improve data processing efficiency and application breadth, and meet the needs of multiple scenarios such as emergency monitoring and target recognition and extraction. Summary of the Invention

[0007] This invention solves the technical problem that existing technologies rely on the deployment of ground control points and subsequent manual registration, resulting in complex processing procedures that make it difficult to meet the needs of rapid response and large-scale data collection.

[0008] The UAV aerial image georegistration method of the present invention includes the following steps:

[0009] Step 1: Use a drone to take aerial photos of the target area and obtain several drone aerial images. Read the information of each drone aerial image.

[0010] Step 2: Based on the information from the UAV aerial images, calculate the geographic coordinates of several image feature points in each UAV aerial image, and use these image feature points as initial control points.

[0011] Step 3: Perform feature extraction and matching on the several UAV aerial images described in Step 1 to obtain robust feature point pairs, and establish a set of common view feature points among all UAV aerial images based on the obtained robust feature point pairs.

[0012] Step 4: Based on the initial control points and the set of common view feature points, construct a multi-view joint optimization model, and use the multi-view joint optimization model to optimize the camera attitude parameters, camera spatial position parameters and geographic coordinates of the initial control points of each UAV aerial image.

[0013] Step 5: Using the optimized camera attitude parameters, optimized camera spatial position parameters, and optimized geographic coordinates of the initial control points, fit a geometric transformation model to perform geometric correction and resampling on the UAV aerial images, and obtain georegistered UAV aerial images.

[0014] Furthermore, in one embodiment of the present invention, the drone aerial image information in step 1 includes the latitude and longitude of the center point at the time of shooting, flight altitude, camera focal length, sensor size, image resolution, and drone camera attitude parameters.

[0015] Furthermore, in one embodiment of the present invention, step 2, calculating the initial geographic coordinates corresponding to several image feature points in each UAV aerial image, includes the following steps:

[0016] Step 21: Select the center point and four corner points of the drone aerial image as image feature points;

[0017] Step 22: Transform the pixel coordinates of the image feature points to the camera coordinate system;

[0018] Step 23: Construct a rotation matrix based on the information from the UAV aerial imagery to transform the vectors of image feature points in the camera coordinate system to the direction vectors in the geographic coordinate system;

[0019] Step 24: Based on the information from the UAV aerial imagery, calculate the ground coordinates corresponding to the image feature points through spatial projection, and use them as the initial geographic coordinates.

[0020] Furthermore, in one embodiment of the present invention, the feature extraction and matching in step 3 includes the following steps:

[0021] Step 31: Extract feature points from each drone aerial image and calculate feature point descriptors;

[0022] Step 32: Perform feature point matching based on feature point descriptors to obtain a candidate matching point set;

[0023] Step 33: Remove mismatched points from the candidate matching point set to obtain robust feature point pairs.

[0024] Furthermore, in one embodiment of the present invention, the multi-view joint optimization model in step 4 adopts a global reprojection error function, introduces a weight and constraint mechanism into the multi-view joint optimization model, and iteratively solves the multi-view joint optimization model. In each iteration, the camera attitude parameters, camera spatial position parameters and geographic coordinates of the initial control point are updated synchronously until the preset convergence condition is met.

[0025] Furthermore, in one embodiment of the present invention, the type of geometric transformation model in step 5 is adaptively selected based on the number of control points, the area size of the UAV aerial image, and the degree of deformation of the UAV aerial image. The type includes affine transformation, quadratic polynomial transformation, or thin plate spline transformation.

[0026] Furthermore, in one embodiment of the present invention, the number of drone aerial images in step 1 is 2 to 6.

[0027] The UAV aerial image georegistration system of the present invention is constructed based on the above method and includes the following modules:

[0028] The reading module uses a drone to take aerial photos of the target area, obtains several drone aerial images, and reads the information of each drone aerial image.

[0029] The calculation module calculates the geographic coordinates of several image feature points in each drone aerial image based on the information of the drone aerial image, and uses the several image feature points as initial control points.

[0030] The feature extraction and matching module performs feature extraction and matching on several UAV aerial images obtained by the reading module to obtain robust feature point pairs, and establishes a set of common view feature points among all UAV aerial images based on the obtained robust feature point pairs.

[0031] The optimization module constructs a multi-view joint optimization model based on the initial control points and the common view feature point set. The multi-view joint optimization model is used to optimize the camera attitude parameters, camera spatial position parameters and geographic coordinates of the initial control points of each UAV aerial image.

[0032] The correction module uses optimized camera attitude parameters, optimized camera spatial position parameters, and optimized geographic coordinates of the initial control points to fit a geometric transformation model, and performs geometric correction and resampling on the UAV aerial images to obtain georegistered UAV aerial images.

[0033] The computer program product described in this invention includes a computer program or instructions that, when executed by a processor, implement the above-described UAV aerial image georeferencing method.

[0034] This invention solves the technical problems of existing technologies that rely on the deployment of ground control points and subsequent manual registration, resulting in complex processing procedures and difficulty in meeting the needs of rapid response and large-scale data acquisition. Specific beneficial effects include:

[0035] This invention proposes a georegistration method for UAV aerial imagery. Instead of focusing on the traditional construction of high-precision orthophotos, it aims to achieve low-redundancy automatic georegistration of UAV aerial imagery for emergency and rapid acquisition scenarios. Traditional methods, when used for emergency response without planned flight paths, high-precision POS (positioning and orientation system), and relying solely on a limited number (2-6) of randomly captured images, suffer from insufficient overlap between images. This significantly reduces the number of effective feature points available for feature matching, resulting in highly sparse feature matching relationships between images. It is difficult to construct a stable and connected multi-view geometric constraint model, making it challenging to correctly initialize the camera pose estimation process based on multi-view geometry. Consequently, the 3D reconstruction process often fails to converge properly or even fails outright. To address the aforementioned technical challenges, this invention employs initial spatial constraint modeling based on aerial EXIF ​​parameters. Without requiring the reconstruction of a complete 3D scene model, it jointly optimizes camera attitude parameters, camera spatial position parameters, and a small number of control points to minimize global reprojection error. This method is simple, requires no complex preprocessing or rigorous flight path planning, and offers significant speed improvements. Furthermore, in emergency response and rapid large-scale data acquisition scenarios, its positioning accuracy can reach sub-meter levels, fully meeting the accuracy and timeliness requirements of emergency response and rapid monitoring. This represents a practical breakthrough from "unable to locate" to "rapidly obtaining usable geographic references." Attached Figure Description

[0036] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0037] Figure 1 This is a flowchart of the UAV aerial image georegistration method described in Implementation Method 1. Detailed Implementation

[0038] Various embodiments of the present invention will now be clearly and completely described with reference to the accompanying drawings. The embodiments described with reference to the drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0039] Implementation Method 1: Existing UAV image georegistration methods based on multi-view geometric reconstruction typically rely on multiple aerial images with high overlap (e.g., no less than 60%) and relatively accurate initial camera pose information. Geographical registration is achieved through image feature point matching, camera pose estimation, and 3D reconstruction. This approach, based on high-precision POS or planned flight paths, heavily depends on expensive specialized equipment and rigorous pre-planning. However, in emergency response or rapid data acquisition scenarios, UAVs often lack pre-planned flight paths, resulting in a limited number of aerial images—for example, only 2–6 randomly captured images with irregular shooting angles, significant differences in front and back viewpoints, and an overlap area of ​​only 10%–30%. Therefore, traditional methods may not be feasible in sudden emergencies or scenarios without specialized equipment.

[0040] To address the aforementioned technical problems, this embodiment proposes a georegistration method for UAV aerial imagery, such as... Figure 1 As shown, this method utilizes a single UAV launch to acquire a small number (2–6) of UAV aerial images at a relatively high altitude. It does not rely on flight path planning or high-overlap acquisition. Initial control points are constructed through projection calculations of the center point and image corner points, providing initial geospatial references for subsequent registration. Without restoring the complete 3D scene model, only camera attitude parameters and a small number of control points are jointly optimized to minimize reprojection errors of common-view feature points. By reducing the number of images and simplifying optimization variables and geometric models, minute-level image stitching and georegistration results can be output. In emergency response and rapid large-scale data acquisition scenarios, its positioning accuracy can reach sub-meter level, meeting the timeliness requirements of emergency response and rapid monitoring, and is suitable for emergency response and rapid large-scale data acquisition scenarios.

[0041] The method for georeferencing drone aerial images includes the following steps:

[0042] Step 1: Use a drone to take aerial photos of the target area and obtain several drone aerial images. Read the information of each drone aerial image.

[0043] Read the EXIF ​​metadata from the drone aerial imagery to obtain the center point latitude and longitude, flight altitude, camera focal length, sensor size, image resolution, and drone camera attitude parameters at the time of shooting.

[0044] EXIF information is a standard data format for recording shooting conditions and location in drone aerial imagery.

[0045] Step 2: Based on the information from the UAV aerial images, calculate the geographic coordinates of several image feature points in each UAV aerial image, and use these image feature points as initial control points.

[0046] Step 3: Perform feature extraction and matching on the several UAV aerial images described in Step 1 to obtain robust feature point pairs, and establish a set of common view feature points among all UAV aerial images based on the obtained robust feature point pairs.

[0047] By analyzing the matching relationships among all UAV aerial images, common view feature points appearing across multiple UAV aerial images are clustered into a "common view feature point set". Common view feature points can provide global constraints for subsequent multi-view joint optimization, including the multiple observation pixel positions of common view feature points in different images and the attitude and positional relationships of different images. Each common view feature point stores the pixel coordinates of multiple images, the UAV aerial image ID, and the corresponding initial control point coordinates. These common view points will serve as input data for multi-view joint optimization (Bundle Adjustment).

[0048] Step 4: Based on the initial control points and the set of common view feature points, construct a multi-view joint optimization model, and use the multi-view joint optimization model to optimize the camera attitude parameters, camera spatial position parameters and geographic coordinates of the initial control points of each UAV aerial image.

[0049] Step 5: Using the optimized camera attitude parameters, optimized camera spatial position parameters, and optimized geographic coordinates of the initial control points, fit a geometric transformation model to perform geometric correction and resampling on the UAV aerial images, and obtain georegistered UAV aerial images.

[0050] Using the unified exterior orientation elements (camera pose parameters), camera spatial position parameters, and geographic coordinates of the initial control points (corner points, center points) obtained by multi-view joint optimization in step 4, a mapping relationship between pixel space and geographic space is constructed, thereby completing geometric correction.

[0051] This implementation includes control point construction, geometric transformation model fitting, and resampling.

[0052] Points observed in at least three images are selected from common viewpoints and scored based on observation count, feature intensity, initial reprojection error, and local density. The image is divided into an N×N grid, and within each grid, points are selected based on their scores while satisfying minimum pixel spacing constraints. Simultaneously, image center points and corner control points are forcibly retained. Finally, the points are truncated to the target number based on their scores. During the geometric transformation fitting process, different weights are assigned to the control points based on their reprojection error or stability index.

[0053] After multi-view joint optimization is completed, the image coordinates of the center point and corner control points of each image are obtained under a unified geographic reference frame, and their corresponding real geographic coordinates are also obtained from the aerial photography parameters and optimization results.

[0054] Therefore, this implementation constructs the following set of control points:

[0055] ;

[0056] in, These are the pixel coordinates in the image. For the corresponding geographic coordinates, This is the number of control points, typically greater than 5. This set of control points is used to fit a geometric transformation model from pixel space to geographic space.

[0057] After completing the geometric model fitting, a mapping function from the target geographic space (i.e., the projection space) to the original pixel space is constructed. The image is resampled using inverse mapping and interpolation to generate an image with geospatial reference.

[0058] After geometric correction, the system saves the imagery as a standard format file (GeoTIFF format) with geospatial reference information. The file embeds projection parameters and geographic coordinates, enabling geospatial positioning. Additionally, it outputs control point coordinate files for easy import and further analysis into GIS software. Batch processing capabilities are supported, improving the overall efficiency of UAV aerial imagery from acquisition to spatial applications.

[0059] Using the method described in this embodiment, when conducting drone aerial photography operations over an area of ​​approximately 1 square kilometer, no pre-planned flight path is required. With the drone flying at an altitude of 400 meters and the image overlap rate controlled at 30%, only about 6 drone aerial images are needed to cover the target area. The corresponding data acquisition time is approximately 5 minutes, and the data processing time is approximately 5 minutes. This allows for the acquisition of geographically referenced aerial images in a short time, and the positioning accuracy meets the needs of applications such as emergency response and rapid monitoring.

[0060] In contrast, traditional UAV aerial surveying methods using planned flight routes typically require flight planning and multiple flight paths to acquire data over an area of ​​approximately 1 square kilometer, with the overall flight operation time exceeding 30 minutes. Furthermore, due to the need to collect a large number of highly overlapping images, the subsequent data processing based on multi-view geometric reconstruction usually takes 3–5 hours to complete under common computing configurations (e.g., i7 processor, 32GB RAM). Therefore, traditional methods are insufficient in terms of operational timeliness and processing efficiency to meet the demands of emergency scenarios for rapidly acquiring usable geographic information.

[0061] Implementation Method 2: This implementation method further defines the UAV aerial image georegistration method described in Implementation Method 1. Step 2, which calculates the initial geographic coordinates corresponding to several image feature points in each UAV aerial image, includes the following steps:

[0062] Step 21: Select the center point and four corner points of the drone aerial image as image feature points;

[0063] Step 22: Transform the pixel coordinates of the image feature points to the camera coordinate system;

[0064] Step 23: Construct a rotation matrix based on the information from the UAV aerial imagery to transform the vectors of image feature points in the camera coordinate system to the direction vectors in the geographic coordinate system;

[0065] Step 24: Based on the information from the UAV aerial imagery, calculate the ground coordinates corresponding to the image feature points through spatial projection, and use them as the initial geographic coordinates.

[0066] This implementation method calculates the geographic coordinates corresponding to the four corner points and the center point in the UAV aerial image based on the camera parameters (including focal length, sensor size, and image resolution) extracted from the UAV aerial image, as well as the flight altitude and camera attitude parameters of the UAV. These coordinates serve as initial geospatial reference points for subsequent multi-view joint optimization input of the UAV aerial image.

[0067] 1. Camera model and pixel coordinate normalization

[0068] Convert pixels in the drone aerial imagery (e.g., top left, top right, bottom left, bottom right, and center point) into normalized image planar coordinates. Given a drone aerial imagery resolution of... × pixels, camera focal length (Unit: mm), the dimensions of the sensor in the horizontal and vertical directions are as follows: , (Unit: mm), then the physical size of each pixel is:

[0069] ;

[0070] Let the pixel coordinates be The main point (center point) of the drone aerial image is ( , ) = ( / 2, / 2), then its normalized coordinates in the camera coordinate system are:

[0071] ;

[0072] ;

[0073] ;

[0074] That is, the vector of this pixel in the camera coordinate system is:

[0075] ;

[0076] 2. Coordinate rotation transformation (camera attitude parameters)

[0077] Based on camera attitude parameters (heading angle) during drone aerial photography Pitch angle Roll angle Construct three Euler angle rotation matrices:

[0078] ;

[0079] ;

[0080] ;

[0081] Synthetic camera pose rotation matrix:

[0082] ;

[0083] Transform the vector in the camera coordinate system into a direction vector in the geographic coordinate system:

[0084] ;

[0085] 3. Solving for the intersection of spatial projection and ground.

[0086] Let the spatial location of the drone during aerial photography be . This refers to the latitude, longitude, and flight altitude of the center point of the drone aerial image. The coordinates of the control point are obtained by extending the direction vector in the geographic coordinate system and intersecting it with the ground plane (assuming it is a plane here).

[0087] Without high-precision terrain data, the ground can be approximated as a plane. If terrain elevation data is available, the terrain elevation can be used to replace the planar model, and the model can be scaled accordingly.

[0088] ;

[0089] in, The elevation value of the ground point, ( , , ) is the ray direction vector pointing from the UAV's position to the ground target point.

[0090] The projected positions (X, Y) of the image corner points in the geographic coordinate system are calculated. Finally, combined with the latitude and longitude of the center point, the corresponding latitude and longitude coordinates are obtained by geographic projection inverse calculation (WGS-84 inverse calculation in this case).

[0091] 4. Output control point coordinate set

[0092] Repeat the above process to calculate the geographic coordinates of five control points, including the center point and corner points of the image. Use these as initial geographic coordinates for subsequent joint optimization input of UAV aerial images.

[0093] Implementation Method 3: This implementation method further defines the UAV aerial image georegistration method described in Implementation Method 1. Step 3, feature extraction and matching, includes the following steps:

[0094] Step 31: Extract feature points from each drone aerial image and calculate feature point descriptors;

[0095] Step 32: Perform feature point matching based on feature point descriptors to obtain a candidate matching point set;

[0096] Step 33: Remove mismatched points from the candidate matching point set to obtain robust feature point pairs.

[0097] This implementation performs feature point extraction and matching on all input UAV aerial images. During feature extraction, the SIFT (Scale Invariant Feature Transform) algorithm is used to extract feature points from each UAV aerial image, and feature point descriptors are calculated. SIFT features are scale, rotation, and illumination invariant, making them suitable for UAV aerial images taken at different attitudes and altitudes. Initial matching involves using a KD-tree (K-dimensional spatial partitioning tree) or FLANN (Fast Nearest Neighbor Search) to search for nearest neighbor pairs of feature vectors, obtaining a candidate matching point set. RANSAC (Random Sample Consensus Algorithm) is used to fit the fundamental matrix (for scenes with elevation differences) or homography matrix (for scenes with flat ground) to filter the candidate matching point set, removing mismatched points and retaining stable and reliable matching pairs, ultimately obtaining robust feature correspondences between each pair of overlapping images.

[0098] Implementation Method 4: This implementation method further defines the UAV aerial image georegistration method described in Implementation Method 1. It utilizes initial control points and common-view feature points to solve for the most consistent camera attitude parameters, control point coordinates, and image projection relationship through a global joint optimization approach.

[0099] For all images and their common viewpoints, a multi-view joint optimization model is constructed, which employs a global reprojection error function:

[0100] ;

[0101] in,( Let be the geographic coordinates of the i-th common viewpoint. Let this be the observed pixel position of the point in the j-th image. Let be the camera pose and position parameters for the j-th image, and let projection(...) be the camera imaging projection parameters used to project spatial points onto the image plane according to the camera parameters.

[0102] The global reprojection error function reflects the deviation between the actual observation and the predicted projection. Its main purpose is to minimize the global reprojection error to achieve geometric consistency between all images, feature points, and camera parameters.

[0103] After completing image feature matching and establishing co-view relationships, this implementation method uses a nonlinear least squares multi-view joint optimization method based on the Levenberg-Marquardt (LM) algorithm to jointly solve the camera parameters and control point geographic coordinates of the UAV aerial images.

[0104] 1. Optimize variable settings

[0105] In the joint optimization process, the camera extrinsic parameters of each UAV aerial image are used as the main optimization objects, including camera attitude parameters and camera spatial position parameters. Simultaneously, the geographic coordinates of control points such as the image center point and image corner points, obtained from the initial estimation of the UAV aerial EXIF ​​parameters, are also included in the optimization variables. When the camera calibration parameters are known and reliable, the camera intrinsic parameters are fixed. When there is some uncertainty in the camera intrinsic parameters, weak constraints are applied to them and they participate in the optimization to improve the overall solution stability.

[0106] 2. Weighting and Constraint Mechanisms

[0107] To enhance the numerical stability and reliability of the optimization process, a weighting and constraint mechanism is introduced into the reprojection error term. Feature points with more common observations and higher matching stability in multiple images are given higher weights, while feature points with larger initial reprojection errors or weaker geometric stability have their weights reduced. At the same time, weak geographic constraints are applied to the image center point and image corner control points generated based on aerial EXIF ​​parameters to suppress the drift of the overall solution and maintain the consistency of geographic reference.

[0108] 3. Iterative solution and result output

[0109] The joint optimization is performed iteratively, synchronously updating camera attitude parameters, camera spatial position parameters, and control point geographic coordinates in each iteration to gradually reduce the global reprojection error. The optimization process terminates when the decrease in global reprojection error falls below a preset threshold, the parameter update amount is less than a convergence threshold, or the maximum number of iterations is reached.

[0110] Through the above joint optimization, globally consistent camera attitude and spatial position parameters for each aerial image are obtained, as well as the initial control point geographic coordinates after joint optimization correction, providing a unified and stable geometric basis for subsequent control point-based geometric correction and geographic registration.

[0111] Implementation Method 5: This implementation method further defines the UAV aerial image georegistration method described in Implementation Method 1. In step 5, the type of geometric transformation model is adaptively selected based on the number of control points, the size of the UAV aerial image region, and the degree of deformation of the UAV aerial image. The type includes affine transformation, quadratic polynomial transformation, or thin plate spline transformation.

[0112] Based on the degree of image distortion and the number of control points, this implementation method adopts a hierarchical strategy to select a suitable geometric transformation model.

[0113] Affine transformation is a linear transformation that preserves the flatness and proportional relationships in an image, making it suitable for registration of UAV aerial images over relatively flat areas. Its two-dimensional spatial transformation model can be expressed as:

[0114] ;

[0115] in, The geographic coordinates of the image. These are the pixel coordinates in the original image. These are the affine transformation parameters. The parameters are obtained by fitting three or more control points using the least squares method.

[0116] When the image distortion is complex or the covered area has certain undulations, a quadratic polynomial transformation is used to enhance the fitting ability. Its model expression is as follows:

[0117] ;

[0118] ;

[0119] in, , For the geographic target coordinates, , These are the pixel coordinates in the image. ...... Let be the coefficients of the quadratic polynomial in the x-direction. ...... Let be the coefficients of the quadratic polynomial in the y-direction.

[0120] When the number of control points is sufficient and uniformly distributed, a thin plate spline (TPS) function can be used for nonlinear geometric correction to further improve the local deformation fitting capability. The spline transformation achieves globally smooth and locally accurate registration by minimizing the bending energy function. Its transformation function can be expressed as:

[0121] ;

[0122] in, , The input image contains pixel coordinates. For the coordinates of the control points, , , For the global affine coefficients, The local deformation weighting coefficient for the control points. For basis functions, commonly used are , This is the Euclidean distance from the current point to the control point.

[0123] This implementation method automatically selects the transformation model based on the number of control points, the size of the image area, and the degree of deformation.

[0124] With fewer control points and less image variation, affine transformation is preferred.

[0125] With appropriate control points and some image distortion, a quadratic polynomial transformation is used.

[0126] For applications with abundant control points and a high degree of precision, spline function transformation is recommended.

[0127] All model parameters are solved or fitted using the least squares method to minimize control point matching errors and ensure the geometric consistency and accuracy of the transformed image.

[0128] Implementation Method Six: A UAV Aerial Image Georegistration System, which is constructed based on the method described in Implementation Method One, and includes the following modules:

[0129] The reading module uses a drone to take aerial photos of the target area, obtains several drone aerial images, and reads the information of each drone aerial image.

[0130] The calculation module calculates the geographic coordinates of several image feature points in each drone aerial image based on the information of the drone aerial image, and uses the several image feature points as initial control points.

[0131] The feature extraction and matching module performs feature extraction and matching on several UAV aerial images obtained by the reading module to obtain robust feature point pairs, and establishes a set of common view feature points among all UAV aerial images based on the obtained robust feature point pairs.

[0132] The optimization module constructs a multi-view joint optimization model based on the initial control points and the common view feature point set. The multi-view joint optimization model is used to optimize the camera attitude parameters, camera spatial position parameters and geographic coordinates of the initial control points of each UAV aerial image.

[0133] The correction module uses optimized camera attitude parameters, optimized camera spatial position parameters, and optimized geographic coordinates of the initial control points to fit a geometric transformation model, and performs geometric correction and resampling on the UAV aerial images to obtain georegistered UAV aerial images.

[0134] Implementation Method Seven: A computer program product, comprising a computer program or instructions, which, when executed by a processor, implements the above-described UAV aerial image georeferencing method.

[0135] The above provides a detailed description of the UAV aerial image georegistration method, system, and product proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for georegistration of UAV aerial images, characterized in that, Includes the following steps: Step 1: Use a drone to take aerial photos of the target area and obtain several drone aerial images. Read the information of each drone aerial image. Step 2: Based on the information from the UAV aerial images, calculate the geographic coordinates of several image feature points in each UAV aerial image, and use these image feature points as initial control points. Step 3: Perform feature extraction and matching on the several UAV aerial images described in Step 1 to obtain robust feature point pairs, and establish a set of common view feature points among all UAV aerial images based on the obtained robust feature point pairs. Step 4: Based on the initial control points and the set of common view feature points, construct a multi-view joint optimization model, and use the multi-view joint optimization model to optimize the camera attitude parameters, camera spatial position parameters and geographic coordinates of the initial control points of each UAV aerial image. Step 5: Using the optimized camera attitude parameters, optimized camera spatial position parameters, and optimized geographic coordinates of the initial control points, fit a geometric transformation model to perform geometric correction and resampling on the UAV aerial images, and obtain georegistered UAV aerial images.

2. The UAV aerial image georegistration method according to claim 1, characterized in that, The information for each drone aerial image in step 1 includes the latitude and longitude of the center point at the time of shooting, flight altitude, camera focal length, sensor size, image resolution, and drone camera attitude parameters.

3. The UAV aerial image georegistration method according to claim 1, characterized in that, Step 2, which calculates the geographic coordinates of several image feature points in each UAV aerial image, includes the following steps: Step 21: Select the center point and four corner points of the drone aerial image as image feature points; Step 22: Transform the pixel coordinates of the image feature points to the camera coordinate system; Step 23: Construct a rotation matrix based on the information from the UAV aerial imagery to transform the vectors of image feature points in the camera coordinate system to the direction vectors in the geographic coordinate system; Step 24: Based on the information from the UAV aerial imagery, calculate the ground coordinates corresponding to the image feature points through spatial projection, and use them as the initial geographic coordinates.

4. The UAV aerial image georegistration method according to claim 1, characterized in that, The feature extraction and matching in step 3 includes the following steps: Step 31: Extract feature points from each drone aerial image and calculate feature point descriptors; Step 32: Perform feature point matching based on feature point descriptors to obtain a candidate matching point set; Step 33: Remove mismatched points from the candidate matching point set to obtain robust feature point pairs.

5. The UAV aerial image georegistration method according to claim 1, characterized in that, In step 4, the multi-view joint optimization model adopts a global reprojection error function. A weight and constraint mechanism is introduced into the multi-view joint optimization model to iteratively solve the multi-view joint optimization model. In each iteration, the camera attitude parameters, camera spatial position parameters and geographic coordinates of the initial control point are updated synchronously until the preset convergence condition is met.

6. The UAV aerial image georegistration method according to claim 1, characterized in that, In step 5, the type of geometric transformation model is adaptively selected based on the number of control points, the size of the UAV aerial image region, and the degree of deformation of the UAV aerial image. The types include affine transformation, quadratic polynomial transformation, or thin plate spline transformation.

7. The UAV aerial image georegistration method according to claim 1, characterized in that, In step 1, the number of drone aerial images ranges from 2 to 6.

8. A UAV aerial image georegistration system, wherein the system is constructed based on the UAV aerial image georegistration method of claim 1, characterized in that, Includes the following modules: The reading module uses a drone to take aerial photos of the target area, obtains several drone aerial images, and reads the information of each drone aerial image. The calculation module calculates the geographic coordinates of several image feature points in each drone aerial image based on the information of the drone aerial image, and uses the several image feature points as initial control points. The feature extraction and matching module performs feature extraction and matching on several UAV aerial images obtained by the reading module to obtain robust feature point pairs, and establishes a set of common view feature points among all UAV aerial images based on the obtained robust feature point pairs. The optimization module constructs a multi-view joint optimization model based on the initial control points and the common view feature point set. The multi-view joint optimization model is used to optimize the camera attitude parameters, camera spatial position parameters and geographic coordinates of the initial control points of each UAV aerial image. The correction module uses optimized camera attitude parameters, optimized camera spatial position parameters, and optimized geographic coordinates of the initial control points to fit a geometric transformation model, perform geometric correction and resampling on the UAV aerial images, and obtain georegistered UAV aerial images.

9. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by the processor, they implement the UAV aerial image georeferencing method according to any one of claims 1-7.

Images