Control point automatic matching method for aerial scene
By using an automatic control point matching method for aerial photography scenes, combined with SIFT feature matching and global motion recovery structure optimization algorithm, the problem of control point matching in UAV aerial photography is solved, and efficient and accurate control point positioning is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- VITU INTELLIGENT TECHNOLOGY (SHENZHEN) CO LTD
- Filing Date
- 2025-08-26
- Publication Date
- 2026-06-19
Smart Images
Figure CN121074151B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of aerial image processing technology, specifically a method for automatic control point matching in aerial photography scenes. Background Technology
[0002] In UAV aerial image localization tasks, it is necessary to obtain the precise location of each aerial photograph, including latitude and longitude, altitude, and rotation information (pitch angle, roll angle, and yaw angle). To achieve this, it is usually necessary to combine several control points to perform localization correction on the aerial photographs. A control point is a three-dimensional coordinate point in the real world, located on the ground or a building. For each control point, its geographical coordinate information (longitude, latitude, and altitude) is accurately determined using RTK high-precision positioning equipment. The localization correction method is as follows: for each control point, it is necessary to manually mark the point in all aerial photographs where it can be observed to obtain precise pixel coordinates. Then, by combining the geographical coordinates of the control point with its pixel coordinates in the aerial photographs, a localization optimization algorithm is used to perform localization correction on the UAV aerial photographs, thus obtaining the precise location of each aerial photograph.
[0003] Under the current technological background, the main difficulties in matching control points are as follows: First, measuring the coordinates of control points consumes a lot of manpower and resources, especially when the aerial image area is located in a remote, sparsely populated area. Second, the matching process requires manual point setting. For large aerial areas with a large number of control points, the pixel coordinates of each control point need to be determined from all the observed photos, which is a lot of work. If control points have already been established in the aerial area in the past, it is very difficult to manually match accurate control points from existing aerial photos as the terrain changes. Summary of the Invention
[0004] This invention addresses the above-mentioned problems by proposing an automatic control point matching method for aerial photography scenes. The method aims to automatically match control points based on aerial photography photos, historical control points, historical aerial photos, and positioning results, thereby avoiding the high labor intensity of manual point setting.
[0005] To achieve the above objectives, the present invention provides an automatic control point matching method for aerial photography scenes, the method comprising the following steps:
[0006] The aerial photos in the historical control point map are aligned with the aerial photos of the control points to be matched to obtain the aerial photos after the ground resolution is aligned.
[0007] Feature extraction is performed on historical control point maps;
[0008] Feature extraction is performed on aerial photographs that require control point matching.
[0009] Calculate pose and optimize camera intrinsic parameters for aerial photographs that require control point matching;
[0010] For control point observation photos, aerial photos in the historical control point map are directly filtered based on the observation information between control points and aerial photos in the historical control point map. For aerial photos of the control point that need to be matched, the geographic coordinates of the control point are transformed to the geocentric coordinate system. The poses of all aerial photos of the control point that need to be matched, calculated in the pose calculation and camera intrinsic information optimization steps, are also transformed to the geocentric coordinate system. Combined with the camera intrinsic information of each photo, projection is performed. If the control point can be projected onto the pixel plane of the photo, then the photo is added to the control point's observation photos. For all control points, their corresponding observation photos are filtered.
[0011] For all observation photos, the SIFT feature matching operator is used to perform pairwise feature matching for all observation photos. Geometric constraints are then used to filter out mismatches in the SIFT matching results to obtain feature matching information. For all control points, pairwise feature matching is performed on the observation photos.
[0012] Preliminary control point matching: Based on the matching information obtained from the feature matching step of the observed photos, the information is input into the global motion recovery structure optimization algorithm to obtain several sparse geographic point information. The sparse geographic point, along with all the corresponding observation information and pixel coordinates, is added to the matching result of the current historical control points to obtain preliminary control point matching information. Each sparse geographic point corresponds to the pixel coordinate position in several aerial photos, and there is one and only one pixel position in each photo.
[0013] The control point matching information is organized and filtered. For historical control points with sparse geographic point matching information, the pixel coordinates of all aerial photos of the control points to be matched are merged. If the number of aerial photos of the control points to be matched in the merged result is less than 2, then the historical control point is determined to have failed to match. If the number of aerial photos of the control points to be matched is greater than or equal to 2, then the historical control point is determined to have matched successfully. For all historical control points, the control point matching information is organized and filtered.
[0014] The restoration of historical control point matching information involves, for all historical control points, outputting the name of the aerial photograph of the control point to be matched and its pixel coordinates within the photograph for those control points that meet the matching conditions during the control point matching information processing and filtering. If the size of the aerial photograph of the control point to be matched has been adjusted during the ground resolution alignment step between the aerial photograph in the historical control point map and the aerial photograph of the control point to be matched, the coordinate information must be restored to the original photograph size corresponding to the original data of the aerial photograph of the control point to be matched.
[0015] Furthermore, the resolution alignment method specifically involves: calculating the value of a / b=c between the ground resolution 'a' of the aerial photograph in the historical map and the ground resolution 'b' of the aerial photograph of the control point to be matched;
[0016] If the value of c is between 0.85 and 1.15, no adjustment is needed;
[0017] If the c value is less than 0.85, it is determined that the size of the ground objects in the aerial photos of the historical control point map is larger than the size of the ground objects in the aerial photos of the control points to be matched. In this case, the size of all aerial photos in the historical control point map is reduced by a / b. The width and height information, pixel focal length and imaging principal point values in the camera intrinsic information of all aerial photos in the historical control point map are also reduced by a / b. The pixel coordinate values of each control point in the historical control point map in all aerial photos that can be observed are also reduced by a / b.
[0018] If the value of c is greater than 1.15, the size of ground objects in the aerial photos of the control points to be matched is larger than the size of ground objects in the aerial photos of the historical control point map. Therefore, the size of all aerial photos of the control points to be matched needs to be reduced by a reduction ratio of b / a. The width and height information, pixel focal length and imaging principal point values in the camera intrinsic information of all aerial photos of the control points to be matched also need to be reduced by a reduction ratio of b / a.
[0019] Furthermore, the method for feature extraction from historical control point maps specifically involves: for each aerial photograph in a historical control point map, using the SIFT feature extraction operator to extract features from locations containing pixel coordinates of any control point; for other areas that do not contain pixel coordinates of control points, using the SIFT feature detection operator to detect feature points, and using the SIFT feature extraction operator to extract SIFT features from the detected feature points.
[0020] Furthermore, the method for feature extraction of aerial photos that need to be matched with control points is as follows: for each aerial photo, the SIFT feature detection operator is used to detect feature points, and then the SIFT feature operator is used to extract SIFT features from the detected feature points.
[0021] Furthermore, the method for calculating pose and optimizing camera intrinsic information for aerial photographs requiring control point matching is as follows: For any aerial photograph, the matching pair selection strategy is as follows: First, based on the GPS information of the photograph, select photographs that are geographically close to it for matching; then, compare them using the bag-of-words model and select photographs with similar semantics to supplement the matching pair; for each matched image, use the SIFT feature matching operator for feature matching, and combine geometric constraints to filter out mismatches in the SIFT matching results; input the filtered mismatch information into the global motion recovery structure optimization algorithm for calculation to determine the pose of the photograph requiring control point matching; and obtain the optimized camera intrinsic information for each photograph.
[0022] Furthermore, the control point is a three-dimensional point that includes longitude, latitude, and altitude information; and the three-dimensional point also includes the precise pixel coordinate position information in all observation images corresponding to the control point.
[0023] Furthermore, the aerial photographs are RGB color images of the ground taken by the aircraft during flight.
[0024] Furthermore, the pose refers to the position and attitude information of the camera when the aerial photo was taken. The position information includes longitude, latitude, and altitude information; the attitude information includes pitch angle, roll angle, and yaw angle information.
[0025] Furthermore, the camera intrinsic parameter information describes the parameters of the camera in the pinhole camera model, including the image width and height information w and h, pixel focal lengths fx and fy, imaging principal points cx and cy, and distortion parameters k1, k2, k3, p1, p2.
[0026] Furthermore, the control point map describes a number of control points established in a historical time period for the drone aerial photography area, a number of aerial photos, pose information corresponding to the aerial photos, camera intrinsic parameter information corresponding to the aerial photos, and pixel coordinate information of the control points in all aerial photos that can be observed.
[0027] Compared with existing technologies, the automatic control point matching method for aerial photography scenes provided by this invention eliminates the workflow of manually determining the coordinates of control points in aerial photographs that require control point matching. By using sparse geographic points obtained through a global motion recovery structure optimization algorithm to filter control point matching information, the matching results are more accurate, avoiding mismatches of control points during manual point setting. Attached Figure Description
[0028] Figure 1 The image provided in this application is a partial image of the original control point map data of the historical control point map.
[0029] Figure 2 The embodiments provided in this application contain a set of 362 local images of aerial photographs that have been collected and require matching control points.
[0030] Figure 3 This is a schematic diagram of SIFT feature point matching for an embodiment provided in this application.
[0031] Figure 4 A schematic diagram of control points successfully matched for the embodiments provided in this application. Detailed Implementation
[0032] This embodiment provides a method for automatically matching control points in aerial photography scenes, the method including the following steps:
[0033] S100 performs ground resolution alignment between aerial photos in the historical control point map and aerial photos of the control points to be matched, and obtains aerial photos after ground resolution alignment.
[0034] Because the flight altitudes of the aerial photos in the historical control point map and the aerial photos of the control points to be matched are different, as are the intrinsic parameters of the aerial cameras used, the ground resolution in the images is inconsistent. Therefore, it is necessary to align the ground resolutions.
[0035] S200 performs feature extraction on historical control point maps;
[0036] The S300 extracts features from aerial photographs that require matching control points.
[0037] The S400 calculates pose and optimizes camera intrinsic parameters for aerial photographs that require matching control points.
[0038] S500 Control Point Observation Photo Screening: For each control point in the historical control point map, observation photos are screened. These photos fall into two categories: the first category refers to aerial photographs from the historical control point map, which are screened directly based on the observation information between the control point and the aerial photograph; the second category refers to aerial photographs from which the control point can be observed, screened by transforming the geographic coordinates of the control point to the Geocentric Geofixed (ECEF) coordinate system. In step S400, the poses of all aerial photographs from the control points to be matched, calculated in the pose calculation and camera intrinsic information optimization step, are also transformed to the ECEF coordinate system. Combined with the camera intrinsic information of each photograph, projection is performed, which can be done using a pinhole camera model. If the control point can be projected onto the pixel plane of the photograph, then that photograph is added to the control point's observation photos. For all control points, their corresponding observation photos are screened.
[0039] In step S600, feature matching of observation photos is performed. For each historical control point, observation photos have already been selected in step S500. Features of observation photos belonging to the historical control point map have been extracted in step S200, and features of observation photos belonging to the aerial photographs of the control points to be matched have been extracted in step S300. For all observation photos, pairwise feature matching is performed using the SIFT feature matching operator, and geometric constraints (fundamental matrix / essential matrix) are used to filter out mismatches in the SIFT matching results to obtain feature matching information. For all control points, pairwise feature matching is performed on the observation photos.
[0040] S700 Initial control point matching: For each historical control point, the matching information obtained in step S600 based on the feature matching of the observed photos is input into the Global Structure From Motion (GEM) algorithm to obtain several sparse geographic point information. For each sparse geographic point, if it can be observed by both the observed photos belonging to the historical control point map and the aerial photos of the control point to be matched, and the number of observed photos belonging to the historical control point map is greater than one, and the pixel coordinates in the observed photos belonging to the historical control point map completely match the observed coordinates of the current historical control point in the corresponding photos, then the sparse geographic point, along with all corresponding observation information and pixel coordinates, is added to the matching result of the current historical control point to obtain preliminary control point matching information. Here, each sparse geographic point corresponds to a pixel coordinate position in several aerial photos, and each photo contains exactly one pixel position; the several aerial photos are a subset of the observed photos corresponding to the historical control point.
[0041] S800 Control point matching information processing and filtering: For the control point matching information initially obtained in step S700, for each historical control point, there may be no sparse geographic points or there may be several sparse geographic points. Each sparse geographic point has at least one pixel coordinate in the aerial photograph of the control point to be matched. For historical control points with sparse geographic point matching information, the pixel coordinates in the aerial photographs of all control points to be matched are merged. If multiple different pixel coordinates appear after merging for the same photograph, the average pixel distance between the multiple pixel coordinates is calculated. If it is less than 2 pixels, the average of the above pixel coordinates is taken as the historical control point observation point for matching the aerial photograph; otherwise, the photograph and all corresponding pixel observation coordinates are directly deleted. Then, the summary result corresponding to the historical control point is counted. If the number of aerial photographs of the control point to be matched in the summary result is less than 2, then the historical control point is determined to have failed to match. If the number of aerial photographs of the control point to be matched is greater than or equal to 2, then the historical control point is determined to have matched successfully. For all historical control points, the control point matching information is processed and filtered.
[0042] S900 Restoration of Historical Control Point Matching Information: For all historical control points, for aerial photos that meet the matching conditions with the control point to be matched during the control point matching information sorting and filtering in step S800, output the name of the aerial photo of the control point to be matched and its pixel coordinate information in the photo. If the size of the aerial photo of the control point to be matched has been adjusted during the ground resolution alignment step between the aerial photos in the historical control point map and the aerial photos of the control point to be matched, the coordinate information needs to be restored to the photo size corresponding to the original data of the aerial photos of the control point to be matched.
[0043] Furthermore, the resolution alignment method specifically involves: calculating the value of a / b=c between the ground resolution 'a' of the aerial photograph in the historical map and the ground resolution 'b' of the aerial photograph of the control point to be matched;
[0044] If the value of c is between 0.85 and 1.15, no adjustment is needed;
[0045] If the c value is less than 0.85, it is determined that the size of the ground objects in the aerial photos of the historical control point map is larger than the size of the ground objects in the aerial photos of the control points to be matched. In this case, the size of all aerial photos in the historical control point map is reduced by a / b. The width and height information, pixel focal length and imaging principal point values in the camera intrinsic information of all aerial photos in the historical control point map are also reduced by a / b. The pixel coordinate values of each control point in the historical control point map in all aerial photos that can be observed are also reduced by a / b.
[0046] If the value of c is greater than 1.15, the size of ground objects in the aerial photos of the control points to be matched is larger than the size of ground objects in the aerial photos of the historical control point map. Therefore, the size of all aerial photos of the control points to be matched needs to be reduced by a reduction ratio of b / a. The width and height information, pixel focal length and imaging principal point values in the camera intrinsic information of all aerial photos of the control points to be matched also need to be reduced by a reduction ratio of b / a.
[0047] Furthermore, the method for feature extraction from historical control point maps specifically involves: for each aerial photograph in a historical control point map, using the SIFT feature extraction operator to extract features from locations containing pixel coordinates of any control point; for other areas that do not contain pixel coordinates of control points, using the SIFT feature detection operator to detect feature points, and using the SIFT feature extraction operator to extract SIFT features from the detected feature points.
[0048] Furthermore, the method for feature extraction of aerial photos that need to be matched with control points is as follows: for each aerial photo, the SIFT feature detection operator is used to detect feature points, and then the SIFT feature operator is used to extract SIFT features from the detected feature points.
[0049] Furthermore, the method for calculating pose and optimizing camera intrinsic information for aerial photographs requiring control point matching is as follows: In step S300, since feature points have been extracted from the photographs, feature point matching is first required. For any aerial photograph, the matching pair selection strategy is as follows: first, based on the GPS information of the photograph, select photographs that are geographically close to it for matching; then, compare them using the bag-of-words model and select photographs with similar semantics to add to the matching pair; for each matching pair image, use the SIFT feature matching operator for feature matching, and combine geometric constraints (fundamental matrix / essential matrix) to filter mismatches in the SIFT matching results; input the filtered mismatched information into the Global Structure From Motion (GSM) algorithm for calculation to determine the pose of the photograph requiring control point matching; and obtain the optimized camera intrinsic information for each photograph.
[0050] By using the above methods and steps, the workflow of manually determining the coordinates of control points in aerial photographs that require control point matching is eliminated through automatic control point matching.
[0051] By using sparse geographic points obtained through the global motion recovery structure optimization algorithm to filter control point matching information, the control point matching results are more accurate, avoiding mismatches of control points during the manual point-setting process.
[0052] Furthermore, the control point is a three-dimensional point that includes longitude, latitude, and altitude information; and the three-dimensional point also includes the precise pixel coordinate position information in all observation images corresponding to the control point.
[0053] Furthermore, the aerial photographs are RGB color images of the ground taken by the aircraft during flight.
[0054] Furthermore, the pose refers to the position and attitude information of the camera when the aerial photo was taken. The position information includes longitude, latitude, and altitude information; the attitude information includes pitch angle, roll angle, and yaw angle information.
[0055] Furthermore, the camera intrinsic parameter information describes the parameters of the camera in the pinhole camera model, including the image width and height information w and h, pixel focal lengths fx and fy, imaging principal points cx and cy, and distortion parameters k1, k2, k3, p1, p2.
[0056] Furthermore, the control point map describes a number of control points established in a historical time period for the drone aerial photography area, a number of aerial photos, pose information corresponding to the aerial photos, camera intrinsic parameter information corresponding to the aerial photos, and pixel coordinate information of the control points in all aerial photos that can be observed.
[0057] Please refer to Figures 1 to 4 The following is an implementation of control point matching using the method provided by this invention:
[0058] 1. We now have historical control point map data, which includes 279 aerial photographs, such as... Figure 1The image shown is a partial view of the map of this control point. Each image's pose is composed of rotation and position information. Taking image 0afbb812b14bd9614dca782b1107dbb5707d18d1.jpg as an example, its ID is "0_6", which is unique. The position information is "longitude":114.12213816404332,"latitude":22.9734481133363,"altitude":190.3271473201245, representing longitude, latitude, and altitude respectively. The rotation information is "pitch":-89.7327666943906,"yaw":-8.865155788125287,"roll":1.1749085819739093, corresponding to pitch, yaw, and roll angles respectively. The camera intrinsic parameters for this dataset are as follows: "width":8192,"height":5456,"fx":12825.6767578125,"fy":12825.6767578125,"cx":4094.7982015074813,"cy":2727.8465660366055,"k1":0,"k2":0,"k3":0,"p1":0,"p2":0, corresponding to the pixel width, pixel height, x-direction pixel focal length, y-direction pixel focal length, x-direction principal point, y-direction principal point, distortion parameters k1,k2,k3,p1,p2, respectively. The camera projection model is a standard pinhole model.This historical control point map contains 1200 control points. Each control point has an ID, location information, and its pixel position on the observed image. For example, control point "0_0" has the ID "0_0" and the location information is "longitude":114.1227392336963,"latitude":22.972282280444194,"altitude":29.454782627812875, corresponding to the control point's longitude, latitude, and altitude, respectively. All observed information is: "observers":{"0_0":{"x":3430.4248,"y":5 296.3125},"0_62":{"x":2383.5874,"y":3659.38599},"0_82":{"x":77 44.95947,"y":1224.98193},"0_127":{"x":3735.25366,"y":2728.3967 3},"0_195":{"x":4020.75488,"y":146.542511},"0_248":{"x":7673.6 4795,"y":3834.42529},"0_270":{"x":2461.70483,"y":997.581665}}. In the "observers" field, "0_0" indicates that the control point was observed by the image with index 0_0, where the pixel coordinates of the point in the x-direction are 3430.4248 and the pixel coordinates in the y-direction are 5296.3125; the meanings of the other fields are similar.
[0059] 2. There is now a collection of 362 aerial photographs that need to be matched with control points, with some details as follows: Figure 2 As shown, the aerial photography area corresponding to this photo overlaps with the aerial photography area corresponding to the original data of the historical control point map in section 1. It also corresponds to a camera intrinsic parameter: "width":6144.0,"height":4096.0,"fx":6649.35058594,"fy":6649.35058594,"cx":3072.0,"cy":2048.0,"k1":0.0,"k2":0.0,"k3":0.0,"p1":0.0,"p2":0.0, which respectively correspond to the pixel width, pixel height, x-direction pixel focal length, y-direction pixel focal length, x-direction principal point, y-direction principal point, and distortion parameters k1, k2, k3, p1, p2. The camera projection model is a standard pinhole model. Now, it is necessary to match this with the historical control point map to determine the control points.
[0060] 3. Align the historical control point map with the aerial photographs of the control points to be matched based on their ground resolution (GSD). Calculations show that the ground resolution in the historical control point map is 0.0137304 meters, while the ground resolution of the aerial photographs of the control points to be matched is 0.0264534 meters. Therefore, the size of the aerial photographs in the historical control point map needs to be reduced, with the image width and height each reduced to 0.519 times their original values. Similarly, the camera intrinsic parameters "width", "height", "fx", "fy", "cx", and "cy" should also be reduced to 0.519 times their original values. This yields the historical control point map with aligned ground resolution, and the aerial photographs of the control points to be matched with this aligned ground resolution are the original data for the aerial photographs of the control points to be matched.
[0061] 4. After locating the control points using the SfM algorithm in the aerial photographs, for each of the 1200 historical control points, photos showing the control point in both the historical control point map and the aerial photographs of the control point to be matched are selected. Then, the SIFT feature point matching algorithm is used to perform feature point matching on the extracted features. A schematic diagram of SIFT feature point matching is shown below. Figure 3 As shown.
[0062] 5. For each historical control point, after matching its observed photographs and performing calculations using the SfM algorithm, the pixel coordinates of the successfully matched historical control points in the aerial photographs of the control points to be matched can be obtained. Since the aerial photographs of the control points to be matched have not been scaled, the obtained pixel coordinates can be used as the matching results of the control points. For 1200 historical control points, a total of 567 were successfully matched. A schematic diagram of the successfully matched control points is shown below. Figure 4 As shown.
Claims
1. A method for automatic matching of control points of aerial scenes, characterized in that, The method includes the following steps: The aerial photos in the historical control point map are aligned with the aerial photos of the control points to be matched to obtain the aerial photos after the ground resolution is aligned. Feature extraction is performed on historical control point maps; Feature extraction is performed on aerial photographs that require control point matching. Calculate pose and optimize camera intrinsic parameters for aerial photographs that require control point matching; For control point observation photos, aerial photos in the historical control point map are directly filtered based on the observation information between control points and aerial photos in the historical control point map. For aerial photos of the control point that need to be matched, the geographic coordinates of the control point are transformed to the geocentric coordinate system. The poses of all aerial photos of the control point that need to be matched, calculated in the pose calculation and camera intrinsic information optimization steps, are also transformed to the geocentric coordinate system. Combined with the camera intrinsic information of each photo, projection is performed. If the control point can be projected onto the pixel plane of the photo, then the photo is added to the control point's observation photos. For all control points, their corresponding observation photos are filtered. For all observation photos, the SIFT feature matching operator is used to perform pairwise feature matching for all observation photos. Geometric constraints are then used to filter out mismatches in the SIFT matching results to obtain feature matching information. For all control points, pairwise feature matching is performed on the observation photos. Preliminary control point matching involves inputting the matching information obtained from the feature matching step of the observed photos into the global motion recovery structure optimization algorithm to obtain information on several sparse geographic points. For each sparse geographic point, if it can be observed by both the observed photos belonging to the historical control point map and the aerial photos of the control point to be matched, and the number of observed photos belonging to the historical control point map is greater than one, and the pixel coordinates in the observed photos belonging to the historical control point map completely match the observed coordinates of the current historical control point in the corresponding photos, then the sparse geographic point, along with all corresponding observation information and pixel coordinates, is added to the matching result of the current historical control point to obtain preliminary control point matching information. Each sparse geographic point corresponds to a pixel coordinate position in several aerial photos, with one and only one pixel position in each photo. The control point matching information is organized and filtered. For historical control points with sparse geographic point matching information, the pixel coordinates of all aerial photos of the control points to be matched are merged. If the number of aerial photos of the control points to be matched in the merged result is less than 2, then the historical control point is determined to have failed to match. If the number of aerial photos of the control points to be matched is greater than or equal to 2, then the historical control point is determined to have matched successfully. For all historical control points, the control point matching information is organized and filtered. The restoration of historical control point matching information involves, for all historical control points, outputting the name of the aerial photograph of the control point to be matched and its pixel coordinates within the photograph for those control points that meet the matching conditions during the control point matching information processing and filtering. If the size of the aerial photograph of the control point to be matched has been adjusted during the ground resolution alignment step between the aerial photograph in the historical control point map and the aerial photograph of the control point to be matched, the coordinate information must be restored to the original photograph size corresponding to the original data of the aerial photograph of the control point to be matched.
2. The method of claim 1, wherein, The resolution alignment method is as follows: the ground resolution 'a' of the aerial photograph in the historical map and the ground resolution 'b' of the aerial photograph of the control point to be matched are used to calculate the value of a / b=c. If the value of c is between 0.85 and 1.15, no adjustment is needed; If the c value is less than 0.85, it is determined that the size of the ground objects in the aerial photos of the historical control point map is larger than the size of the ground objects in the aerial photos of the control points to be matched. In this case, the size of all aerial photos in the historical control point map is reduced by a / b. The width and height information, pixel focal length and imaging principal point values in the camera intrinsic information of all aerial photos in the historical control point map are also reduced by a / b. The pixel coordinate values of each control point in the historical control point map in all aerial photos that can be observed are also reduced by a / b. If the value of c is greater than 1.15, the size of ground objects in the aerial photos of the control points to be matched is larger than the size of ground objects in the aerial photos of the historical control point map. Therefore, the size of all aerial photos of the control points to be matched needs to be reduced by a reduction ratio of b / a. The width and height information, pixel focal length and imaging principal point values in the camera intrinsic information of all aerial photos of the control points to be matched also need to be reduced by a reduction ratio of b / a.
3. The method of claim 1, wherein, The specific method for feature extraction from historical control point maps is as follows: for each aerial photograph in a historical control point map, for locations containing pixel coordinates of any control point, the SIFT feature extraction operator is used for feature extraction; for other areas that do not contain pixel coordinates of control points, the SIFT feature detection operator is used for feature point detection, and for the detected feature points, the SIFT feature extraction operator is used for SIFT feature extraction.
4. The method of claim 1, wherein, The method for feature extraction of aerial photos that need to be matched with control points is as follows: for each aerial photo, the SIFT feature detection operator is used to detect feature points, and then the SIFT feature operator is used to extract SIFT features from the detected feature points.
5. The method of claim 1, wherein, The method for calculating pose and optimizing camera intrinsic information for aerial photographs requiring control point matching is as follows: For any aerial photograph, the matching pair selection strategy is as follows: First, based on the GPS information of the photograph, select photographs that are geographically close to it for matching; then, compare them using the bag-of-words model and select photographs with similar semantics to add to the matching pair; for each matching pair image, use the SIFT feature matching operator for feature matching, and combine geometric constraints to filter out mismatches in the SIFT matching results; input the information after filtering out mismatches into the global motion recovery structure optimization algorithm for calculation to determine the pose of the photograph requiring control point matching; and obtain the optimized camera intrinsic information for each photograph.
6. The method of claim 1, wherein, The control point is a three-dimensional point that includes longitude, latitude, and altitude information; and the three-dimensional point also includes the precise pixel coordinate position information in all the observation images corresponding to the control point.
7. The method of claim 1, wherein, The aerial photographs are RGB color images of the ground taken by the aircraft during flight.
8. The method of claim 1, wherein, The pose refers to the position and attitude information of the camera when the aerial photo was taken. The position information includes longitude, latitude, and altitude information; the attitude information includes pitch angle, roll angle, and yaw angle information.
9. The method of claim 1, wherein, The camera intrinsic parameters describe the parameters of the camera in the pinhole camera model, including the image width and height information w and h, pixel focal lengths fx and fy, imaging principal points cx and cy, and distortion parameters k1, k2, k3, p1, p2.
10. The method of claim 1, wherein, The control point map describes a number of control points established in a historical time period for a drone aerial photography area, a number of aerial photos, pose information corresponding to the aerial photos, camera intrinsic parameter information corresponding to the aerial photos, and pixel coordinate information of the control points in all aerial photos that can be observed.