Low-altitude remote sensing image processing method and device, storage medium and computer device

Abstract

The application discloses a low-altitude remote sensing image processing method and device, a storage medium and computer equipment, and mainly aims at improving the processing, service and application efficiency of low-altitude aerial remote sensing images. The method comprises the following steps: using computing power resources to sequentially perform a series of operations such as preprocessing, regional network construction, aerial triangulation, dense point extraction, color and light uniformity, logical inlay and the like on low-altitude aerial remote sensing images in different spatial positions and different scales, and determining a geometric orientation parameter set of the low-altitude aerial remote sensing images in different spatial positions and different scales based on operation results; in response to an orthographic correction signal of the low-altitude aerial remote sensing image, obtaining processing requirement information of an image demand user, wherein the processing requirement information comprises a target spatial position and a target scale; determining a target orientation parameter in the geometric orientation parameter set based on the processing requirement information, and performing orthographic correction processing on the low-altitude aerial remote sensing image with the target orientation parameter.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a low-altitude remote sensing image processing method and device, a storage medium and a computer device. BACKGROUND

[0002] With the continuous development of low-altitude aviation technology, especially the rapid development of low-altitude economy, low-altitude aerial photography technology is also widely used. The amount of low-altitude aerial remote sensing image data is growing explosively, which brings great challenges to the rapid processing, storage and application of remote sensing data.

[0003] At present, each time the image orthorectification is performed, the image needs to be processed through the whole process again. However, this method is time-consuming and laborious, resulting in low efficiency of remote sensing image application service, and the traditional processing mode cannot carry out dynamic production according to the actual needs of users, and cannot meet the customized needs of customers. SUMMARY

[0004] The present application provides a low-altitude remote sensing image processing method, device, storage medium and computer device, which can improve the efficiency and quality of image data processing and application service.

[0005] According to a first aspect of the present application, a low-altitude aerial remote sensing image processing method is provided, comprising:

[0006] A series of operations of preprocessing, regional network construction, aerial triangulation, dense point extraction, color and light uniformity, and logical mosaic of low-altitude aerial remote sensing images at different spatial positions and different scales are performed using computing resources, and a set of geometric orientation parameters of the low-altitude aerial remote sensing images at different spatial positions and different scales is determined based on the operation results;

[0007] In response to a low-altitude aerial remote sensing image orthorectification signal, the processing requirement information of the image requirement user is obtained, wherein the processing requirement information includes a target spatial position and a target scale;

[0008] Based on the processing requirement information, the target orientation parameter is determined in the set of geometric orientation parameters, and the low-altitude aerial remote sensing image with the target orientation parameter is orthorectified.

[0009] Optionally, the preprocessing of the low-altitude aerial remote sensing image at different spatial positions and different scales includes:

[0010] Downsampling processing of the low-altitude aerial remote sensing image at different scales is performed, based on the downsampling processing results at different scales, an image pyramid is constructed, and based on the spatial range of the low-altitude aerial remote sensing image, each layer of image in the image pyramid is block compressed and stored to obtain low-altitude aerial remote sensing images at different spatial positions and different scales.

[0011] Optionally, orthorectification processing is performed on the low-altitude aerial remote sensing image with the target directional parameter, including:

[0012] Obtaining an image distortion correction parameter, dynamically correcting pixel point coordinates in the low-altitude aerial remote sensing image with the target directional parameter by using a preset distortion correction model, and determining the low-altitude aerial remote sensing image after distortion correction from the corrected pixel point coordinates;

[0013] Constructing a projection grid point triangulation for a target ground area covered by the low-altitude aerial remote sensing image after distortion correction;

[0014] Determining a barycentric coordinate of a triangulation to which each grid pixel point to be corrected in the low-altitude aerial remote sensing image after distortion correction belongs, and determining an elevation value corresponding to each grid pixel point to be corrected based on the barycentric coordinate;

[0015] Determining a sensor parameter corresponding to the low-altitude aerial remote sensing image after distortion correction, and performing projection transformation on each grid pixel point to be corrected by using a preset projection transformation algorithm based on the sensor parameter and the elevation value, to obtain a low-altitude aerial remote sensing image after preliminary correction.

[0016] Optionally, after performing projection transformation on each grid pixel point to be corrected by using a preset projection transformation algorithm to obtain a low-altitude aerial remote sensing image after preliminary correction, the method further includes:

[0017] Determining a texture coordinate of each vertex in a triangulation to which each pixel point to be processed in the low-altitude aerial remote sensing image after preliminary correction belongs, determining a triangulation area of the triangulation and a sub-triangulation area of a sub-triangulation corresponding to the triangulation, and taking a ratio of the triangulation area to the sub-triangulation area as a texture barycentric coordinate;

[0018] Determining a corrected texture coordinate corresponding to each pixel point to be processed based on the texture barycentric coordinate and the texture coordinate, and performing texture correction on each pixel point to be processed based on the corrected texture coordinate, to obtain a low-altitude aerial remote sensing image after orthorectification.

[0019] Optionally, determining an elevation value corresponding to each grid pixel point to be corrected based on the barycentric coordinate includes:

[0020] Taking any vertex in the triangulation as a target vertex, determining a plurality of control points around the target vertex in the target ground area, and determining a distance from each control point to the target vertex and ground elevation information of each control point;

[0021] determine a vertex elevation value of the target vertex based on the distance and the ground elevation information, and determine an elevation value corresponding to each to-be-corrected grid pixel based on the barycentric coordinate and the vertex elevation value.

[0022] Optionally, before orthorectification processing is performed on the low-altitude aerial remote sensing image with the target orientation parameter, the method further comprises:

[0023] determining a plurality of adjacent low-altitude aerial remote sensing images corresponding to the low-altitude aerial remote sensing image with the target orientation parameter, forming a low-altitude aerial remote sensing image sequence from each adjacent low-altitude aerial remote sensing image and the low-altitude aerial remote sensing image with the target orientation parameter, determining a photographing parameter of each image in the low-altitude aerial remote sensing image sequence, determining a photographing center coordinate of each image based on the photographing parameter, and determining a nadir point coordinate of each image based on the photographing center coordinate;

[0024] constructing a nadir point triangle with each nadir point as a vertex, determining perpendicular bisectors of three sides of the nadir point triangle, and determining an intersection point of any two perpendicular bisectors, taking each intersection point as a polygon vertex, and constructing a polygon from the polygon vertices, and determining a target image region covered by the polygon in the low-altitude aerial remote sensing image with the target orientation parameter;

[0025] performing orthorectification processing on the low-altitude aerial remote sensing image with the target orientation parameter, comprising:

[0026] performing orthorectification processing on the target image region with the target orientation parameter.

[0027] Optionally, constructing a regional mesh for the low-altitude aerial remote sensing image, comprising:

[0028] selecting a low-level pyramid image in the image pyramid, extracting feature points in each low-level pyramid image, and performing cross-image matching of the feature points in the same level to obtain a feature point pair;

[0029] determining a rotation matrix and a translation vector between each low-level pyramid image based on the coordinates of each feature point in the feature point pair, and determining an apparatus intrinsic parameter of a sensor corresponding to the low-level pyramid image;

[0030] taking the rotation matrix, the translation vector, and the apparatus intrinsic parameter as initial regional mesh parameters, determining an image resolution of the low-level pyramid image and an original image resolution of the low-altitude aerial remote sensing image, determining a regional mesh parameter adjustment coefficient based on the image resolution and the original image resolution, and adjusting the initial regional mesh parameters based on the regional mesh parameter adjustment coefficient, and determining a regional mesh of the low-altitude aerial remote sensing image based on the parameter adjustment result.

[0031] According to a second aspect of the present application, a processing device for low-altitude aerial remote sensing images is provided, comprising:

[0032] a parameter determination unit configured to sequentially perform at least one of preprocessing in different spatial positions and different scales, regional net construction, aerial triangulation, dense point extraction, and color and light uniformization on the low-altitude aerial remote sensing images, and determine a set of geometric orientation parameters of the low-altitude aerial remote sensing images in different spatial positions and different scales based on the operation results;

[0033] an acquisition unit configured to acquire processing requirement information of an image requirement user in response to an orthorectification signal of the low-altitude aerial remote sensing images, wherein the processing requirement information comprises a target spatial position and a target scale;

[0034] an orthorectification unit configured to determine a target orientation parameter in the set of geometric orientation parameters based on the processing requirement information, and perform orthorectification processing on the low-altitude aerial remote sensing images with the target orientation parameter.

[0035] According to a third aspect of the present application, a computer readable storage medium is provided, which stores a computer program, and the program is executed by a processor to implement the above processing method for low-altitude aerial remote sensing images.

[0036] According to a fourth aspect of the present application, a computer device is provided, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the above processing method for low-altitude aerial remote sensing images when executing the program.

[0037] The processing method, device, storage medium, and computer device for low-altitude remote sensing images provided by the present application can save the full-process processing time and avoid the situation that an error occurs in a certain process when performing full-process processing, thereby improving the data processing and service efficiency of low-altitude aerial remote sensing images and improving the value of data resources. BRIEF DESCRIPTION OF DRAWINGS

[0038] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:

[0039] Figure 1 A flow chart of a processing method of low-altitude remote sensing images is shown according to an embodiment of the application;

[0040] Figure 2 A flow chart of another processing method of low-altitude remote sensing images is shown according to an embodiment of the application;

[0041] Figure 3 A structural schematic diagram of a processing device of low-altitude remote sensing images is shown according to an embodiment of the application;

[0042] Figure 4 A structural schematic diagram of another processing device of low-altitude remote sensing images is shown according to an embodiment of the application;

[0043] Figure 5 A physical structural schematic diagram of a computer device is shown according to an embodiment of the application. DETAILED DESCRIPTION

[0044] The application will be described in detail below with reference to the drawings and embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0045] Currently, each time orthorectification of images is performed, the image needs to be processed in the whole process again, which is time-consuming and laborious, resulting in low efficiency of remote sensing image application services. Meanwhile, the traditional processing mode cannot carry out dynamic production according to the actual needs of users, and cannot meet the customized needs of customers.

[0046] To solve the above problems, an embodiment of the application provides a processing method of low-altitude remote sensing images, as shown in Figure 1 The method comprises the following steps.

[0047] 101, a series of operations of preprocessing, regional network construction, aerial triangulation, dense point extraction, uniform color and light, and logical inlaying of low-altitude aerial remote sensing images in different spatial positions and different scales are performed by using computing power resources, and a geometric orientation parameter set of the low-altitude aerial remote sensing images in different spatial positions and different scales is determined based on the operation results.

[0048] The low-altitude aerial remote sensing images can be remote sensing images taken by a UAV on any object and region, and the acquisition method and coverage area of the remote sensing images are not limited in the embodiment of the application.

[0049] For the embodiment of the present application, any low-altitude aerial remote sensing image can be preprocessed, regional network constructed, aerial triangulation adjusted, dense point extracted, color and light uniformized, and logically mosaicked in a unified manner before orthorectification, therefore, the embodiment of the present application pre-processes low-altitude aerial remote sensing images in different spatial positions and different scales, constructs regional network, adjusts aerial triangulation, extracts dense points, uniformizes color and light, and logically mosaics, and stores the processed images and their processing parameters as a geometric orientation parameter set, so that when subsequent orthorectification of images in a certain spatial position or scale is needed, the corresponding images can be directly obtained from the geometric orientation parameter set for orthorectification processing, thereby improving the orthorectification efficiency of the images. In order to construct the geometric orientation parameter set, the low-altitude aerial remote sensing image needs to be pre-processed first, based on which, the method comprises: performing down-sampling processing of the low-altitude aerial remote sensing image in different scales, constructing an image pyramid based on the down-sampling processing results in different scales, and performing block compression processing on each layer of image in the image pyramid based on the spatial range of the low-altitude aerial remote sensing image and storing the same, so as to obtain low-altitude aerial remote sensing images in different spatial positions and different scales.

[0050] Specifically, first, the number of pyramid levels is defined, such as 5 levels, then the pyramid of the image is constructed according to the number of levels, for example, if the resolution of each layer is 1, 1 / 2, 1 / 4, 1 / 8, 1 / 16 of the original image in turn, the distorted corrected image is Gaussian down-sampled to generate a hierarchical image pyramid. For example, the resolution of the original image is 4000x3000 pixels; the first layer (original): 4000x3000 pixels; the second layer (original): 2000x1500 pixels; the third layer (original): 1000x750 pixels; the fourth layer (original): 500x375 pixels; the fifth layer (original): 250x188 pixels. Then, according to the longitude and latitude of the image, the image is divided into grids, for each image, the grid blocks covered by the image are determined, for each layer of image, the sub-blocks are generated by cropping according to the grid block boundary, for example, the first layer of image is divided into 40x30 blocks, each block is 100x100 pixels. Finally, each block is compressed and stored to solve the storage space.

[0051] Further, after the image pyramid is constructed, a region network for each image in each pyramid level also needs to be constructed. Based on this, the method comprises: selecting a low-level pyramid image in the image pyramid, and extracting feature points in each low-level pyramid image respectively, and performing cross-image matching of the feature points in the same level to obtain a feature point pair; determining a rotation matrix and a translation vector between each low-level pyramid image based on the coordinates of each feature point in the feature point pair, and determining the device intrinsic parameter of the corresponding sensor of the low-level pyramid image; taking the rotation matrix, the translation vector and the device intrinsic parameter as initial region network parameters, determining the image resolution of the low-level pyramid image and the original image resolution of the low-altitude aerial remote sensing image, determining a region network parameter adjustment coefficient based on the image resolution and the original image resolution, and adjusting the initial region network parameters based on the region network parameter adjustment coefficient, and determining the region network of the low-altitude aerial remote sensing image based on the parameter adjustment result.

[0052] wherein the device intrinsic parameter comprises an intrinsic matrix and a distortion coefficient. Specifically, if the selected low-level pyramid image is a second-level pyramid image, feature points are extracted from each image in the second-level pyramid image by using a method such as SIFT (Scale-Invariant Feature Transform), and cross-image matching of the feature points in the same level is performed to obtain a feature point pair. Then, the coordinates of each feature point in the feature point pair are used to calculate the relative orientation elements (rotation matrix and translation vector) between the images by using a method such as bundle adjustment, and then the relative orientation elements and the device intrinsic parameter are integrated into initial region network parameters. Then, based on the resolution of the second-level pyramid image and the original image resolution, an adjustment coefficient is determined. For example, if the resolution of the second-level pyramid image is 1 / 4 of the original image, the initial region network parameters are multiplied by 4 respectively to obtain the region network of the low-altitude aerial remote sensing image.

[0053] Further, the images after the regional network construction are subjected to spatial triangulation adjustment to find a set of optimal camera parameters (external parameters: position, posture; internal parameters: focal length, principal point, distortion coefficient) and three-dimensional point coordinates, so that the re-projection error of corresponding feature points in all images is minimized, so as to obtain high-precision image internal parameters, external parameters, encrypted points and other results. Further, the images are subjected to dense point extraction, such as matching the same name pixels in different view images, calculating the depth value of each pixel, and then generating a depth map, and finally fusing into a dense point cloud. Then, the images are subjected to color and light uniformity processing, that is, the light difference (such as exposure, color temperature, shadow) between images is eliminated, so that the multi-view images remain consistent in tone and brightness, and the subsequent image orthorectification quality is improved. After all the above processes are completed, the one-time processing result of the images and the parameters obtained by processing are obtained, and the one-time processing result and the parameters obtained by processing constitute a geometric orientation parameter set. That is, the geometric orientation parameter set includes the images after processing in different spatial positions and different scales, and the parameters obtained in the image processing process.

[0054] 102, in response to the low-altitude aerial remote sensing image orthorectification signal, obtaining the processing requirement information of the image requirement user, wherein the processing requirement information includes the target spatial position and the target scale.

[0055] For the embodiment of the application, if the user needs to orthorectify the image of the target spatial position and the target scale, the pre-processed, regional network constructed, aerial triangulation adjusted, dense point extracted, and color and light uniformized image of the target spatial position and the target scale is directly obtained from the geometric orientation parameter set, and the image is directly orthorectified, without the need to re-process the image through the whole process of pre-processing, regional network construction, aerial triangulation adjustment, dense point extraction, and color and light uniformization, so as to improve the orthorectification efficiency of the image.

[0056] 103, determining the target orientation parameter in the geometric orientation parameter set based on the processing requirement information, and orthorectifying the low-altitude aerial remote sensing image with the target orientation parameter.

[0057] For the embodiment of the present application, the pre-processing of obtaining the target space position and the target scale in the geometric orientation parameter set, the region network construction, the aerial triangulation, the dense point extraction, the uniform color and light after the image, before the orthographic correction of the image, in order to further improve the correction efficiency, the target influence area which needs to be corrected is also determined in the image, based on this, the method comprises: determining a plurality of adjacent low-altitude aerial remote sensing images corresponding to the low-altitude aerial remote sensing image with the target orientation parameter, forming a low-altitude aerial remote sensing image sequence by each adjacent low-altitude aerial remote sensing image and the low-altitude aerial remote sensing image with the target orientation parameter, determining the photographic center coordinates of each image in the low-altitude aerial remote sensing image sequence based on the photographic parameters of each image, and determining the image bottom point coordinates of each image based on the photographic center coordinates; constructing an image bottom point triangle by taking each image bottom point as a vertex, determining the perpendicular bisectors of the three sides in the image bottom point triangle, and determining the intersection points of any two perpendicular bisectors, taking each intersection point as a polygon vertex, and constructing a polygon from the polygon vertex, and determining the target image area covered by the polygon in the low-altitude aerial remote sensing image with the target orientation parameter.

[0058] Specifically, the low-altitude aerial remote sensing image sequence I1, I2,..., I n , the photographic parameters such as the camera focal length f, the unmanned aerial vehicle flight height h and the attitude angle (φ, ω, k), the photographic center coordinates C i (X i ,Y i ,Z i ) of each image i are calculated according to the photographic parameters. pi ,Y pi ) are calculated based on the photographic center coordinates as follows:

[0059]

[0060] Where Δx and Δy are the deviations of the image center pixel coordinates in the x direction and the y direction respectively. Thus, the image bottom point coordinates of each image i can be determined in the above manner. Then, the image bottom points are connected into a triangular network which does not contain an obtuse triangle, each triangle is composed of three image bottom points, for each triangle, the perpendicular bisectors L ab , L bc , L ca of the three sides are calculated, the intersection points of the perpendicular bisectors are the vertices of the polygon to be constructed, all the vertices are connected to form the polygon V i of each image bottom point, and the polygon V iThe covered area is taken as a target image area needing orthographic correction, and finally the target image area with a target orientation parameter is subjected to orthographic correction processing. Since when collecting low-altitude aerial remote sensing images, in order to meet the data model processing requirements, the images in a flight strip need to have a certain degree of overlap, and high overlap leads to too much repeated calculation of the area in real-time processing, wasting computing resources, therefore, the embodiment of the present application determines the non-overlapping target image area in the form of a polygon, which can avoid repeated orthographic correction of a certain area, save resources, and improve the orthographic correction efficiency of the image. In the orthographic correction process, the fast scheduling of multi-scale projection grids and textures. According to the user's demand scale, for example, the resolution of the original image is 0.1 meters, and the user currently needs to access the scale of 0.8 meters, then the real-time calculation algorithm can directly read the grid data in the related area of the 3rd layer of the pyramid to meet the user's demand. In actual operation, for example, the user needs to access 0.6 meters (the user's application scale can be continuous, and does not necessarily need to match the scale space of the data body), then according to the access demand (resolution priority, application efficiency priority), 0.4 meters (the 2nd layer of the pyramid) or 0.8 meters (the 3rd layer of the pyramid) can be intelligently selected for real-time calculation and rendering. Image block and projection grid mapping: for a certain area of projection grid, each related image block, according to the algorithm of the 4th step, the size of the projection grid corresponding to the image block of different pyramid levels is also inconsistent, which can be calculated according to the parameters of the constructed model. For example, if the image grid size and the image block are consistent, then each image block of image 0 (the image body) corresponds to a corresponding grid, each image block of image 1 layer pyramid corresponds to 4 image grids, each image block of image 2 layer pyramid corresponds to 16 image grids, and so on. If the image grid size and the image block are 1:2, then each image block corresponds to 4 image grids. In this way, each image block can know the projection grid relationship corresponding to each block. The interpolation resampling of image pixel information is realized by using a computer graphics acceleration device (GPU) for texture rendering. For the dynamic change demand of the user for the image resolution (scale) in the real-time calculation process, a dynamic texture rendering method based on multi-scale (combining multi-scale projection grid and image pyramid) is proposed, realizing the real-time processing of the pixel color information of the orthographic correction image.Texture rendering triangle mesh construction: the read image block data and projection grid data, construct texture object and triangle net, the construction of triangle net follows the consistency principle, for example: 4 grid points A, B, C, D are connected in turn clockwise to form a quadrilateral without intersecting edge, then the triangle ABC, ACD two triangles cover the grid; All the grid can be constructed according to the similar principle, that is, the triangle net of all related projection grid can be constructed; In practical application, the scale of each image block corresponding to the triangle net can be dynamically adjusted according to the scale of user access, instead of constructing triangle net for rendering every time, it can be rendered according to (2, 4, 8…) times, improve the efficiency of triangle net rendering, avoid unnecessary rendering. Image orthographic correction based on GPU rendering: the image block is constructed into a texture object, and then rendered according to the triangle net and texture coordinates to realize the conversion from image grid to projection grid. The rendering result is the orthographic corrected image in the geodetic projection coordinate system, realizing the orthographic correction based on real-time calculation. The process completes the real-time correction of an image file. All images within the user demand range can be processed and rendered in real time according to the above method, and the orthographic image rendering result of all images in the whole region can be obtained.

[0061] According to the low-altitude aerial remote sensing image processing method provided by the application, compared with the method of reprocessing the image in the whole process every time, the application performs at least one of the following processes on all low-altitude aerial remote sensing images in advance: preprocessing of different spatial positions and different scales, regional net construction, aerial triangulation, dense point extraction, and uniform color and light, to determine a set of geometric orientation parameters of the remote sensing image based on the processing result. When orthographic correction of a remote sensing image is needed, the target geometric orientation parameter set of the remote sensing image is obtained directly from the set of geometric orientation parameters according to the user's processing demand for the image, and finally the image is orthographically corrected based on the target geometric orientation parameter set. Thus, the preprocessed image processing parameters are reused to directly perform the subsequent image orthographic correction, which can save the whole process processing time and avoid errors in the whole process, thereby improving the orthographic correction efficiency and accuracy of low-altitude aerial remote sensing images.

[0062] Further, in order to better illustrate the above process of processing low-altitude remote sensing images, as a refinement and extension of the above embodiment, the embodiment of the application provides another method for processing low-altitude remote sensing images, as shown in Figure 2 The method comprises:

[0063] 201. Using computing resources, perform a series of operations on low-altitude aerial remote sensing images at different spatial locations and scales, including preprocessing, regional network construction, aerial triangulation adjustment, dense point extraction, color and light equalization, and logical mosaicking. Based on the operation results, determine the set of geometric orientation parameters of low-altitude aerial remote sensing images at different spatial locations and scales.

[0064] Specifically, (1) Low-altitude aerial remote sensing image data mainly includes sensor parameters, sensor status (position and attitude) at the moment of data acquisition, image data, etc. Preprocessing includes distortion correction and multi-scale pyramid construction. ① Distortion correction in actual operation includes two aspects: static and dynamic correction. The static part can be compensated using the distortion correction parameters from the laboratory factory; the dynamic part is compensated using the results of aerial triangulation adjustment. In actual operation, the image coordinates of the pixel points on the image are calculated according to the distortion model to achieve correction. In the real-time calculation process of this invention, only the image points with the same name generated in the construction of the regional network and the grid image points in the construction of the projection grid need to be compensated and corrected in real time. There is no need to process each pixel, thereby improving the processing efficiency. ② Multi-scale image space construction is a key step that affects and determines the implementation of subsequent real-time processing algorithms. Its core purpose is to achieve fast and efficient reading of image data. To reduce the intensity of image reading, especially file writing, this embodiment of the invention uses an external file storage model for the multi-scale image pyramid. Each pyramid layer is resampled sequentially in a 1:2 ratio. For example, if the original image size is 10240×10240 pixels, the first pyramid layer is 5120×5120 pixels, and so on, with subsequent layers being 2560×2560, 1280×1280, 640×640, 320×320, etc. Each pyramid layer is organized and stored using a block-based grid structure, such as a block size of 128×128. Following this block size, the height and width of the final pyramid layer must be smaller than the block size. In the example above, the final pyramid layer is 80×80 pixels. Furthermore, each pyramid layer block is compressed (using lossy or lossless compression depending on actual needs) for storage, further reducing the intensity of subsequent data I / O and improving I / O efficiency. ③ In practical applications, the original data can be further reorganized and stored in a multi-scale pyramid manner to reduce data storage space and improve data efficiency.

[0065] (2) Multi-scale regional network construction and adjustment key includes two contents of regional network construction and aerial triangulation adjustment. ① The embodiment of the application can realize primary regional network construction by using low-level pyramid data, and on the basis of initial value, high-precision regional network is constructed, and image data I0 is less; in the actual operation process, it is not necessary to construct each level, and only individual level is selected to construct regional network to obtain initial value, and the same name area can be accurately positioned in the bottom layer data by using the initial value, and the search efficiency and success rate of the same name point are improved. ② Aerial triangulation adjustment is a mathematical process of data statistics and calculation solution, and high-precision image internal parameter, external parameter and encryption point are obtained. ③ In the actual application process, the above process only needs to be processed once, and subsequent processing only needs the result of the above process, so in the application process, the original geometric orientation result is generally stored in a database or an additional file, instead of repeated construction and solution, so that the subsequent processing efficiency of the image can be improved. In addition, the results of the above processing can be directly included in the data source library, and directly used in the subsequent instant calculation ortho correction and service.

[0066] (3) Digital elevation model is constructed according to the geometric accuracy requirement of data service. ① If there is a standard digital elevation model result in the data range, it can be directly used; ② If not, the terrain data can be obtained by interpolating and fitting the encryption points, which can meet the emergency service and rapid application of image results; ③ If the existing digital elevation model precision cannot meet the precision requirement, the dense point matching digital surface model is used, the non-ground points are filtered out by using the automatic filtering program, and then the high-precision digital elevation model is obtained. The embodiment of the application gives the processing means and process of digital elevation model in combination with the actual application requirement, and the main purpose is to quickly solve the input data source problem of subsequent real-time ortho correction, and in the actual application process, the spatio-temporal terrain database can also be constructed.

[0067] (4) Construction of the projection grid model. The projection grid model is the key to establishing a corresponding transformation relationship between the image coordinate system and the geodetic coordinate system. This invention uses the regular grid coordinates of regular images as the starting point for projection transformation. During the implementation process, the image grid is kept consistent with the image blocks (or multiples of 2, 4, 8, 16, etc.), which is important for subsequent multi-scale projection grid dynamic texture rendering updates. ① Constructing the image grid. For example, if the image storage block is 128×128 and the image size is 10240×10240, then a grid point is taken every 128 pixels in height and width to construct an 81×81 grid; if the grid is selected to be twice the size of the image block, then the grid interval is 64 pixels, and the image grid is 161×161, and the image coordinates of each grid point can be calculated. ② Calculation of ground coordinates for image grid points: This process can be solved using the collinearity condition equation in photogrammetry. The core principle is the constraint that the image coordinates on the image, the spatial position and orientation of the camera center, and the collinearity of the corresponding ground objects. Since the image coordinates of the image grid points are known, and the position and orientation of the camera center can be obtained, using the collinearity condition equation and the digital elevation model (DEM), the coordinates of the ground objects are the intersection of the line connecting the image point and the camera center with the DEM. In practice, an iterative approximation method can be used to calculate and construct the image grid projected from the image coordinates to the object coordinates. After the above processing is completed, the processed image and its corresponding processing parameters (where the processing parameters can be image spatial position, image resolution, image area grid, color and light uniformity parameters, etc.) are obtained. The processed image and its corresponding processing parameters constitute a set of geometric orientation parameters.

[0068] 202. In response to the orthorectification signal of low-altitude aerial remote sensing image, acquire the processing requirement information of the image user, wherein the processing requirement information includes the target spatial location and target scale.

[0069] 203. Determine the target orientation parameters from the set of geometric orientation parameters based on the processing requirements information.

[0070] Specifically, the processed image of the target spatial location and target scale required by the user and its corresponding orientation parameters are obtained from the set of geometric orientation parameters as target orientation parameters.

[0071] 204. Construct a projection grid triangular network for the target ground area covered by low-altitude aerial remote sensing imagery with target orientation parameters.

[0072] Before constructing the triangular net, distortion correction needs to be performed on the triangular net, based on which the method comprises: acquiring image distortion correction parameters, dynamically correcting pixel point coordinates in the low-altitude aerial remote sensing image with target orientation parameters by using a preset distortion correction model, and determining the low-altitude aerial remote sensing image after distortion correction by using the corrected pixel point coordinates. Then, a projection grid point triangular net is constructed for a target ground area covered by the low-altitude aerial remote sensing image after distortion correction.

[0073] Specifically, the image distortion correction parameters are the transpose intrinsic parameters of the image shooting device, such as camera intrinsic parameters (including intrinsic matrix and distortion coefficients). Specifically, based on the image distortion correction parameters, the preset distortion correction model is used to dynamically correct the image, so as to obtain the low-altitude remote sensing image coordinates after distortion correction. In order to improve the correction accuracy of the preset distortion correction model, the preset distortion correction model needs to be trained and constructed first, based on which the method comprises: constructing a preset initial distortion correction model; acquiring a sample data set, wherein the sample data set comprises sample low-altitude remote sensing images and corresponding corrected images; dividing the sample data set into a training set and a test set, training the preset initial distortion correction model by using the training set, and testing the trained preset initial distortion correction model by using the test set, and finally taking the preset initial distortion correction model that meets the test condition as the preset distortion correction model. Specifically, in the model training process, the preset initial distortion correction model is first constructed, and then the sample data set is acquired. It is ensured that the data set contains all necessary files, including a plurality of sample low-altitude remote sensing images and corresponding corrected images. The data is converted into a format that can be understood by the preset initial distortion correction model, and finally the model is trained and tested. Specifically, the data set can be divided first: the sample data set is divided into a training set and a test set by using a random or specific strategy (such as stratified sampling). Then, the model is trained by using the training set, and the trained model is tested by using the test set to evaluate its performance on unseen data. The mCP, accuracy, recall rate and other indicators on the test set are calculated and recorded. If the model performance does not meet the requirements, more iterations or adjustments can be returned to the training stage. In this way, the preset initial distortion correction model that meets the requirements is obtained.

[0074] In another embodiment of the present application, in the process of dynamically correcting the image, the flight attitude (roll angle, pitch angle, yaw angle) and flight height corresponding to the shooting time of each image can also be acquired, and based on the flight attitude and flight height, the rotation matrix of the image is determined, and the dynamic distortion calibration model parameter is obtained. The embodiment of the present application adopts a static and dynamic combination to correct the image, which can flexibly solve the source of image distortion, improve the geometric correction accuracy, and enhance the adaptability to complex scenes.

[0075] Further, according to the range of the image and the required output resolution, a regular projection grid is defined in a target coordinate system (such as a geodetic coordinate system). For example, the coverage area of the mountainous image is divided into square grids with a side length of 10 meters, and each grid intersection point is a projection grid point. The projection grid points are connected to form a triangular network.

[0076] 205、determine the barycentric coordinates of each pixel point of the low-altitude aerial remote sensing image with the target orientation parameters in the triangular network to which the pixel point belongs, and determine the corresponding elevation value of each pixel point based on the barycentric coordinates.

[0077] For the embodiment of the present application, for a certain pixel point to be corrected, if the coordinates of the three vertices of the triangular network to which the pixel point belongs are A, B, and C, and the coordinate of the pixel point to be corrected is P, the barycentric coordinates (λ1, λ2, λ3) of the triangular network to which the pixel point belongs are calculated according to the following formula:

[0078] P = λ1A + λ2B + λ3C

[0079] λ1 + λ2 + λ3 = 1

[0080] According to the above method, the barycentric coordinates of each pixel point to be corrected can be determined. Then, based on the barycentric coordinates, the corresponding elevation value of each pixel point to be corrected is determined. Based on this, the method comprises: taking any vertex in the triangular network as a target vertex, determining a plurality of control points around the target vertex in the target ground area, determining the distance from each control point to the target vertex, and determining the ground elevation information of each control point; determining the vertex elevation value of the target vertex based on the distance and the ground elevation information; and determining the corresponding elevation value of each pixel point to be corrected based on the barycentric coordinates and the vertex elevation value.

[0081] Specifically, the vertex elevation value Z of each vertex P in the triangular network is determined according to the following formula: p

[0082]

[0083] where i is the i-th control point, n is the total number of control points, di is the distance from the i-th control point to the vertex P, and Zi is the ground elevation information of the i-th control point. pi According to the above method, the vertex elevation value of each vertex in the triangular network can be determined. Then, the corresponding elevation value Z of each pixel point to be corrected is determined according to the following formula: s

[0084] s Z = λ1Z1 + λ2Z2 + λ3Z3​​​​

[0085] wherein, Z1, Z2, Z3 are respectively the elevation values of the three vertices in the triangular mesh to which the grid pixel point pair belongs.

[0086] 206, determine the sensor parameters corresponding to the low-altitude aerial remote sensing image with the target orientation parameters, based on the sensor parameters and the elevation values, use the preset projection transformation algorithm to perform projection transformation on each grid pixel point to be corrected respectively, and obtain the low-altitude aerial remote sensing image after preliminary correction.

[0087] Specifically, based on the sensor parameters (such as the focal length of the camera lens, the image principal point coordinates, etc.) and the elevation values corresponding to the low-altitude aerial remote sensing image with the target orientation parameters, the pixel points on the original image are projected onto the corrected plane by using the projection transformation principle in computer graphics. GPU quickly completes the projection transformation and color interpolation of the pixel points through the hardware accelerated rendering pipeline, and generates the corrected image. For example, for each triangle, GPU can map the texture of the original image to the corrected triangle according to the vertex coordinates and texture coordinates (corresponding to the pixel position on the original image) of the triangle, so as to realize real-time correction and display of the image.

[0088] Further, after the image is corrected in the above manner, the image also needs to be texture corrected. Based on this, the method comprises the following steps: determining the texture coordinates of each vertex in the triangular mesh to which each pixel point to be processed in the low-altitude aerial remote sensing image after preliminary correction belongs, determining the triangular mesh area of the triangular mesh and the sub-triangular mesh area of the sub-triangular mesh corresponding to the triangular mesh, taking the ratio of the triangular mesh area and the sub-triangular mesh area as the texture barycentric coordinates; based on the texture barycentric coordinates and the texture coordinates, determining the corrected texture coordinates corresponding to each pixel point to be processed, and performing texture correction on each pixel point to be processed based on the corrected texture coordinates, to obtain the low-altitude aerial remote sensing image after ortho-rectification.

[0089] Specifically, the pixel coordinates of each vertex in the triangular mesh are taken as the corresponding texture coordinates. For each triangle, the sub-triangular mesh is generated by barycentric coordinate interpolation or subdivision algorithm (such as dividing the triangle into four small triangles) according to the distribution of the internal pixel points, the area of each sub-triangle is calculated, for example, if three sub-triangles are contained, their areas are S sub1 , S sub2 , S sub3 , and the area of the triangular mesh to which a certain pixel point to be processed belongs is S, then the texture barycentric coordinates (λ w1 , λ w2 , λ w3 ) are determined according to the following formula:

[0090]

[0091] Thus, the texture barycentric coordinates of each pixel to be processed can be determined in the above manner, and then the corrected texture coordinates (u, v) are calculated using linear interpolation according to the texture barycentric coordinates and the texture coordinates (u1, v1), (u2, v2), (u3, v3) of the three vertices of the triangular mesh to which the pixel to be processed belongs:

[0092] u = λ w1 u1 + λ w2 u2 + λ w3 u3

[0093] v = λ w1 v1 + λ w2 v2 + λ w3 v3

[0094] Then, the corrected texture coordinates are assigned to each pixel to be processed in the low-altitude aerial remote sensing image, so as to obtain the orthorectified low-altitude aerial remote sensing image. The embodiment of the present application directly utilizes the original image, integrates the processing process into the computing infrastructure, combines computing power and instant computing processing algorithms, and provides services in real time, dynamically and intelligently according to actual needs of users. That is, the embodiment of the present application uses an instant computing method to realize data processing (GPU, CPU, memory, and video memory) at each stage of data processing; the data processing process of the present application is parameterized, and for processes requiring massive computation such as image matching, image adjustment, orientation, and terrain construction, the storage parameters are processed once (the parameters are directly recorded in the original achievement database), and the subsequent processing directly reads the parameters for processing, so as to avoid the influence of these processes on the computing service efficiency of data; the embodiment of the present application is customized according to user needs, and the processing algorithm intelligently carries out data processing according to the needs of users for image services such as scale, precision, space, and time, so as to avoid waste of computing resources. In summary, the embodiment of the present application is an intelligent information data processing and service method for rapid application of low-altitude aerial remote sensing images, which is proposed by integrating current computing-based information service technology.

[0095] According to the low-altitude aerial remote sensing image processing method provided by the application, compared with the current method of reprocessing the image in the whole process each time when the image orthographic correction is performed, the application performs at least one of preprocessing, regional network construction, aerial triangulation adjustment, dense point extraction, and uniform color and light in different spatial positions and different scales on all low-altitude aerial remote sensing images in sequence once, determines a set of geometric orientation parameters of the remote sensing image based on the processing result, obtains the target set of geometric orientation parameters of the remote sensing image in the set of geometric orientation parameters according to the processing requirement of the user when the orthographic correction of the remote sensing image is needed, and finally performs the orthographic correction on the image based on the target set of geometric orientation parameters. Thus, the image orthographic correction processing can be directly performed by reusing the preprocessed image parameters, the whole processing time can be saved, and the error in the whole processing can be avoided, so that the orthographic correction efficiency and accuracy of the low-altitude aerial remote sensing image can be improved.

[0096] Further, as a specific implementation of Figure 1 , the embodiment of the application provides a low-altitude remote sensing image processing device, as shown in Figure 3 , the device comprises a parameter determination unit 31, an acquisition unit 32, and an orthographic correction unit 33.

[0097] The parameter determination unit 31 can be used for performing a series of operations such as preprocessing, regional network construction, aerial triangulation adjustment, dense point extraction, uniform color and light, and logical inlaying on low-altitude aerial remote sensing images in different spatial positions and different scales by using computing resources, and determining a set of geometric orientation parameters of the low-altitude aerial remote sensing images in different spatial positions and different scales based on the operation result.

[0098] The acquisition unit 32 can be used for acquiring the processing requirement information of the image demand user in response to the low-altitude aerial remote sensing image orthographic correction signal, wherein the processing requirement information comprises a target spatial position and a target scale.

[0099] The orthographic correction unit 33 can be used for determining the target orientation parameter in the set of geometric orientation parameters based on the processing requirement information, and performing the orthographic correction processing on the low-altitude aerial remote sensing image with the target orientation parameter.

[0100] In a specific application scenario, in order to preprocess the low-altitude remote sensing image in different spatial positions and different scales, as shown in Figure 4 , the parameter determination unit 31 comprises a downsampling module 311 and a first construction module 312.

[0101] The downsampling module 311 can be configured to perform downsampling processing on the low-altitude aerial remote sensing image at different scales, and construct an image pyramid based on the downsampling processing results at different scales.

[0102] The first construction module 312 can be configured to perform block compression processing on each layer of image in the image pyramid based on a spatial range of the low-altitude aerial remote sensing image, and store the processed image to obtain low-altitude aerial remote sensing images at different spatial positions and different scales.

[0103] In a specific application scenario, in order to perform orthographic correction processing on the low-altitude aerial remote sensing image with the target directional parameter, the orthographic correction unit 33 includes a second construction module 331, a first determination module 332, and a projection transformation module 333.

[0104] The second construction module 331 can be configured to obtain image distortion correction parameters, dynamically correct pixel point coordinates in the low-altitude aerial remote sensing image with the target directional parameter by using a preset distortion correction model, and determine the low-altitude aerial remote sensing image after distortion correction based on the corrected pixel point coordinates.

[0105] The second construction module 331 can also be configured to construct a projection grid point triangulation network for a target ground area covered by the low-altitude aerial remote sensing image after distortion correction.

[0106] The first determination module 332 can be configured to determine a barycentric coordinate of a triangular network to which each to-be-corrected grid pixel point in the low-altitude aerial remote sensing image after distortion correction belongs, and determine an elevation value corresponding to each to-be-corrected grid pixel point based on the barycentric coordinate.

[0107] The projection transformation module 333 can be configured to determine a sensor parameter corresponding to the low-altitude aerial remote sensing image after distortion correction, perform projection transformation on each to-be-corrected grid pixel point by using a preset projection transformation algorithm based on the sensor parameter and the elevation value, and obtain a low-altitude aerial remote sensing image after preliminary correction.

[0108] In a specific application scenario, in order to utilize the preset projection transformation algorithm to respectively perform projection transformation on each to-be-corrected grid pixel point, the projection transformation module 333 can be specifically configured to determine a texture coordinate of each vertex in a triangular mesh to which each to-be-processed pixel point in the low-altitude aerial remote sensing image after preliminary correction belongs, determine a triangular mesh area of the triangular mesh and a sub-triangular mesh area of a sub-triangular mesh corresponding to the triangular mesh, and take a ratio of the triangular mesh area to the sub-triangular mesh area as a texture barycentric coordinate; based on the texture barycentric coordinate and the texture coordinate, determine a corrected texture coordinate corresponding to each to-be-processed pixel point, and perform texture correction on each to-be-processed pixel point based on the corrected texture coordinate, to obtain the low-altitude aerial remote sensing image after orthographic correction.

[0109] In a specific application scenario, in order to respectively determine the elevation value corresponding to each to-be-corrected grid pixel point, the first determination module 332 can be specifically configured to take any vertex in the triangular mesh as a target vertex, determine a plurality of control points around the target vertex in the target ground area, and determine a distance from each control point to the target vertex and ground elevation information of each control point; based on the distance and the ground elevation information, determine a vertex elevation value of the target vertex, and respectively determine the elevation value corresponding to each to-be-corrected grid pixel point based on the barycentric coordinate and the vertex elevation value.

[0110] In a specific application scenario, in order to determine a target image region covered by a polygon in the low-altitude aerial remote sensing image with the target orientation parameter, the device further includes an image reconstruction unit 34.

[0111] The image reconstruction unit 34 can be configured to determine a plurality of adjacent low-altitude aerial remote sensing images corresponding to the low-altitude aerial remote sensing image with the target orientation parameter, form a low-altitude aerial remote sensing image sequence by each adjacent low-altitude aerial remote sensing image and the low-altitude aerial remote sensing image with the target orientation parameter, respectively determine a photographic parameter of each image in the low-altitude aerial remote sensing image sequence, determine a photographic center coordinate of each image based on the photographic parameter, and determine a foot point coordinate of each image based on the photographic center coordinate; construct a foot point triangle by taking each foot point as a vertex, determine perpendicular bisectors of three sides in the foot point triangle, and determine an intersection point of any two perpendicular bisectors, take each intersection point as a polygon vertex, and construct a polygon by the polygon vertex, to determine the target image region covered by the polygon in the low-altitude aerial remote sensing image with the target orientation parameter.

[0112] In a specific application scenario, in order to perform orthographic correction on the image, the orthographic correction unit 33 can be specifically configured to perform orthographic correction processing on the target image region with the target orientation parameter.

[0113] In a specific application scenario, in order to construct a regional net for low-altitude aerial remote sensing images, the parameter determination unit 31 further comprises an extraction module 313, a second determination module 314, and an adjustment module 315.

[0114] The extraction module 313 can be used to select low-level pyramid images in the image pyramid, extract feature points in each low-level pyramid image respectively, and perform cross-image matching of the feature points in the same level to obtain a feature point pair.

[0115] The second determination module 314 can be used to determine a rotation matrix and a translation vector between each low-level pyramid image based on the coordinates of each feature point in the feature point pair, and determine the device intrinsic parameter of the corresponding sensor of the low-level pyramid image.

[0116] The adjustment module 315 can be used to take the rotation matrix, the translation vector, and the device intrinsic parameter as initial regional net parameters, determine the image resolution of the low-level pyramid image and the original image resolution of the low-altitude aerial remote sensing image, determine a regional net parameter adjustment coefficient based on the image resolution and the original image resolution, adjust the initial regional net parameters based on the regional net parameter adjustment coefficient, and determine the regional net of the low-altitude aerial remote sensing image based on the parameter adjustment result.

[0117] It should be noted that other corresponding descriptions of the functions of the low-altitude remote sensing image processing device provided by the embodiments of the present application can be referred to the corresponding descriptions of the method shown in Figure 1 , which will not be described here in detail.

[0118] Based on the above method shown in Figure 1 , accordingly, the embodiments of the present application also provide a computer readable storage medium having a computer program stored thereon, which is executed by a processor to implement the following steps: using computing power resources to sequentially perform a series of operations such as preprocessing, regional net construction, aerial triangulation, dense point extraction, uniform color and light, and logical mosaic on low-altitude aerial remote sensing images in different spatial positions and different scales, and determining a set of geometric orientation parameters of the low-altitude aerial remote sensing images in different spatial positions and different scales based on the operation results; in response to a low-altitude aerial remote sensing image orthorectification signal, obtaining processing requirement information of an image demand user, wherein the processing requirement information includes a target spatial position and a target scale; determining a target orientation parameter in the set of geometric orientation parameters based on the processing requirement information, and performing orthorectification processing on the low-altitude aerial remote sensing image with the target orientation parameter.

[0119] Based on the above method shown in Figure 1 and the method shown in Figure 3The embodiment of the device also provides an entity structure diagram of a computer device, as shown in the figure Figure 5 As shown, the computer device comprises a processor 41, a memory 42, and a computer program stored on the memory 42 and executable on the processor, wherein the memory 42 and the processor 41 are both arranged on a bus 43, and the processor 41 implements the following steps when executing the program: using computing resources to sequentially perform at least one of preprocessing, regional network construction, aerial triangulation, dense point extraction, color and light uniformity, and logical mosaic on low-altitude aerial remote sensing images at different spatial positions and different scales, and determining a geometric orientation parameter set of the low-altitude aerial remote sensing images at different spatial positions and different scales based on operation results; in response to an orthorectification signal of the low-altitude aerial remote sensing images, obtaining processing requirement information of an image demand user, wherein the processing requirement information comprises a target spatial position and a target scale; determining a target orientation parameter in the geometric orientation parameter set based on the processing requirement information, and performing orthorectification processing on the low-altitude aerial remote sensing images with the target orientation parameter.

[0120] Through the technical scheme of the present application, at least one of preprocessing, regional network construction, aerial triangulation, dense point extraction, color and light uniformity is sequentially performed on all low-altitude aerial remote sensing images at different spatial positions and different scales in advance, and a geometric orientation parameter set of the remote sensing images is determined based on the processing results, when orthorectification of a certain remote sensing image is required, the target geometric orientation parameter set of the remote sensing image is directly obtained in the geometric orientation parameter set according to the processing requirement of the image by the user, and finally the image is orthorectified based on the target geometric orientation parameter set, thereby the image orthorectification processing is directly performed by reusing the preprocessed image processing parameters, the full-process processing time can be saved, and the situation that a certain process is incorrectly processed during the full-process processing is avoided, thereby the orthorectification efficiency and accuracy of the low-altitude aerial remote sensing images are improved.

[0121] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present application can be realized by a general computing device, which can be concentrated on a single computing device or distributed on a network composed of multiple computing devices, and optionally, they can be realized by program codes executable by the computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described herein can be executed in different order, or they can be manufactured into individual integrated circuit modules, or multiple modules or steps thereof can be manufactured into a single integrated circuit module. Thus, the present application is not limited to any specific combination of hardware and software.

[0122] The above merely provides the preferred embodiments of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the principles and technical scope of the present application shall fall into the scope of the present application.

Claims

1. A method for processing low-altitude remote sensing images, characterized in that, include: Using computing resources, a series of operations are performed on low-altitude aerial remote sensing images at different spatial locations and scales, including preprocessing, regional network construction, aerial triangulation adjustment, dense point extraction, color and light equalization, and logical mosaicking. Based on the operation results, the set of geometric orientation parameters of the low-altitude aerial remote sensing images at different spatial locations and scales is determined. In response to the orthorectification signal of low-altitude aerial remote sensing imagery, the processing requirement information of the imagery user is acquired, wherein the processing requirement information includes the target spatial location and the target scale; Based on the processing requirements, target orientation parameters are determined from the set of geometric orientation parameters. Multiple adjacent low-altitude aerial remote sensing images corresponding to the low-altitude aerial remote sensing image with the target orientation parameters are determined. A low-altitude aerial remote sensing image sequence is formed by each adjacent low-altitude aerial remote sensing image and the low-altitude aerial remote sensing image with the target orientation parameters. Aerial photography parameters are determined for each image in the low-altitude aerial remote sensing image sequence. The photographic center coordinates of each image are determined based on the aerial photography parameters, and the image base point coordinates of each image are determined based on the photographic center coordinates. An image base point triangle is constructed using each image base point as a vertex. The perpendicular bisectors of the three sides of the image base point triangle are determined, and the intersection of any two perpendicular bisectors is determined. Each intersection point is used as a polygon vertex, and a polygon is constructed from the polygon vertices. The target image area covered by the polygon is determined in the low-altitude aerial remote sensing image with the target orientation parameters. Orthorectification processing is performed on the target image area with the target orientation parameters.

2. The method according to claim 1, characterized in that, Preprocessing of low-altitude aerial remote sensing images at different spatial locations and scales includes: The low-altitude aerial remote sensing images are downsampled at different scales, and an image pyramid is constructed based on the downsampling results at different scales. Based on the spatial range of the low-altitude aerial remote sensing image, each layer of the image pyramid is segmented, compressed, and stored to obtain low-altitude aerial remote sensing images at different spatial locations and scales.

3. The method according to claim 1, characterized in that, Orthorectification processing is performed on low-altitude aerial remote sensing images with target orientation parameters, including: Obtain image distortion correction parameters, use a preset distortion correction model to dynamically correct the pixel coordinates in the low-altitude aerial remote sensing image with target orientation parameters, and determine the distortion-corrected low-altitude aerial remote sensing image by the corrected pixel coordinates. A projection grid triangular network is constructed for the target ground area covered by the distortion-corrected low-altitude aerial remote sensing image. Determine the centroid coordinates of the triangular mesh of the projected grid point to which each grid pixel to be corrected belongs in the low-altitude aerial remote sensing image after distortion correction, and determine the elevation value corresponding to each grid pixel to be corrected based on the centroid coordinates. The sensor parameters corresponding to the distortion-corrected low-altitude aerial remote sensing image are determined. Based on the sensor parameters and the elevation value, a preset projection transformation algorithm is used to perform projection transformation on each grid pixel to be corrected, so as to obtain the preliminarily corrected low-altitude aerial remote sensing image.

4. The method according to claim 3, characterized in that, After performing projection transformation on each pixel of the grid to be corrected using a preset projection transformation algorithm to obtain a preliminarily corrected low-altitude aerial remote sensing image, the method further includes: The texture coordinates of each vertex in the projection grid triangulation of each pixel to be processed in the preliminarily corrected low-altitude aerial remote sensing image are determined. The area of ​​the triangulation of the projection grid triangulation and the area of ​​the sub-triangulation of the corresponding sub-triangulation of the projection grid triangulation are determined. The ratio of the area of ​​the triangulation to the area of ​​the sub-triangulation is used as the texture centroid coordinates. Based on the texture centroid coordinates and the texture coordinates, the corrected texture coordinates corresponding to each pixel to be processed are determined, and texture correction is performed on each pixel to be processed based on the corrected texture coordinates to obtain the orthorectified low-altitude aerial remote sensing image.

5. The method according to claim 3, characterized in that, Based on the centroid coordinates, the elevation value corresponding to each grid pixel to be corrected is determined, including: In the projected grid triangulation, each vertex is taken as a target vertex. Multiple control points around the target vertex are determined in the target ground area, and the distance from each control point to the target vertex is determined, as well as the ground elevation information of each control point is determined. Based on the distance and the ground elevation information, the vertex elevation value of the target vertex is determined, and based on the centroid coordinates and the vertex elevation value, the elevation value corresponding to each grid pixel to be corrected is determined.

6. The method according to claim 2, characterized in that, Regional network construction for low-altitude aerial remote sensing imagery includes: In the image pyramid, select low-level pyramid images, extract feature points in each low-level pyramid image, and perform cross-image matching at the same level to obtain feature point pairs. Based on the coordinates of each feature point in the feature point pair, the rotation matrix and translation vector between each low-level pyramid image are determined, and the device intrinsic parameters of the sensor corresponding to the low-level pyramid image are determined. Using the rotation matrix, the translation vector, and the device intrinsic parameters as initial regional network parameters, the image resolution of the low-level pyramid image and the original image resolution of the low-altitude aerial remote sensing image are determined. Based on the image resolution and the original image resolution, regional network parameter adjustment coefficients are determined, and the initial regional network parameters are adjusted based on the regional network parameter adjustment coefficients. Based on the parameter adjustment results, the regional network of the low-altitude aerial remote sensing image is determined.

7. A processing apparatus for low-altitude remote sensing images, characterized in that, include: The parameter determination unit is used to perform a series of operations on low-altitude aerial remote sensing images at different spatial locations and scales, including preprocessing, regional network construction, aerial triangulation adjustment, dense point extraction, color and light equalization, and logical mosaicking, using computing resources. Based on the operation results, it determines the set of geometric orientation parameters of the low-altitude aerial remote sensing images at different spatial locations and scales. The acquisition unit is used to acquire the processing requirement information of the image user in response to the orthorectification signal of the low-altitude aerial remote sensing image, wherein the processing requirement information includes the target spatial location and the target scale. An orthorectification unit is used to determine target orientation parameters in the geometric orientation parameter set based on the processing requirement information, determine multiple adjacent low-altitude aerial remote sensing images corresponding to the low-altitude aerial remote sensing image with target orientation parameters, form a low-altitude aerial remote sensing image sequence by each adjacent low-altitude aerial remote sensing image and the low-altitude aerial remote sensing image with target orientation parameters, determine the aerial photography parameters of each image in the low-altitude aerial remote sensing image sequence, determine the photo center coordinates of each image based on the aerial photography parameters, and determine the image base point coordinates of each image based on the photo center coordinates; construct an image base point triangle with each image base point as a vertex, determine the perpendicular bisectors of the three sides of the image base point triangle, determine the intersection of any two perpendicular bisectors, use each intersection point as a polygon vertex, construct a polygon with the polygon vertices, and determine the target image area covered by the polygon in the low-altitude aerial remote sensing image with target orientation parameters; Orthorectification is performed on the target image region with target orientation parameters.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images