Orthophoto generation method, device and storage medium based on three-dimensional Gaussian splashing

Abstract

The present invention discloses a method, device and storage medium for generating orthophoto images based on three-dimensional Gaussian splashing, comprising: in response to an orthophoto generation signal of a building area to be measured, obtaining a multi-view image obtained by photographing the building area to be measured from multiple viewing angles, and determining image acquisition parameters of an image acquisition device corresponding to the multi-view image; performing cross-image matching on feature points in each image, and determining feature point pairs in each image based on the matching results; performing three-dimensional reconstruction on regional points in the building area to be measured represented by the feature point pairs based on the feature point pairs and the image acquisition parameters to obtain a three-dimensional point cloud, and performing Gaussian distribution processing on the three-dimensional point cloud to obtain multiple three-dimensional Gaussian ellipsoids; determining a two-dimensional projection plane, and determining depth information of each three-dimensional Gaussian ellipsoid from the two-dimensional projection plane, and splashing each three-dimensional Gaussian ellipsoid into the two-dimensional projection plane based on the depth information to obtain an orthophoto of the building area to be measured.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method, device and storage medium for generating orthophoto images based on three-dimensional Gaussian splattering. Background Art

[0002] With the continuous development of science and technology, spatial geographic information has been used in more and more fields as the spatial basis of national production. At the same time, digital orthophotos with map geometric accuracy and image characteristics have become one of the most important spatial geographic information data in the national economy, national defense construction, social development and people's lives due to their high accuracy, rich information and easy access.

[0003] At present, orthophotos are usually generated through digital elevation models. However, digital elevation models mainly record terrain elevation information, and often cannot accurately express the height and shape information of artificial or natural obstacles such as buildings. As a result, when generating orthophotos, the areas behind these obstacles may not be projected correctly, resulting in non-orthophotos, that is, the generation accuracy of orthophotos is low. Summary of the invention

[0004] The present invention provides a method, device and storage medium for generating orthophotos based on three-dimensional Gaussian splattering, which are mainly capable of improving the generation accuracy of orthophotos.

[0005] According to a first aspect of the present invention, there is provided a method for generating an orthophoto based on three-dimensional Gaussian splashing, comprising:

[0006] In response to an orthophoto generation signal of the measured building area, a multi-view image is obtained by photographing the measured building area from multiple viewing angles, and an image acquisition parameter of an image acquisition device corresponding to the multi-view image is determined;

[0007] Extracting feature points with significant geometric structures in each of the images, performing cross-image matching on the feature points in each of the images, and determining feature point pairs with similar geometric structures in each of the images based on the matching results;

[0008] Based on the feature point pairs and the image acquisition parameters, three-dimensionally reconstruct the regional points in the measured building area represented by the feature point pairs to obtain a three-dimensional point cloud of the measured building area, and perform Gaussian distribution processing on the three-dimensional point cloud to obtain a plurality of three-dimensional Gaussian ellipsoids corresponding to the measured building area;

[0009] Determine the two-dimensional projection plane of the measured building area, and determine the depth information of each three-dimensional Gaussian ellipsoid from the two-dimensional projection plane. Based on the depth information, splash each three-dimensional Gaussian ellipsoid into the two-dimensional projection plane to obtain an orthophoto of the measured building area.

[0010] According to a second aspect of the present invention, there is provided an orthophoto generation device based on three-dimensional Gaussian splashing, comprising:

[0011] an acquisition unit, configured to generate a signal in response to an orthophoto of a detected building area, acquire a multi-view image obtained by photographing the detected building area from multiple viewing angles, and determine an image acquisition parameter of an image acquisition device corresponding to the multi-view image;

[0012] A matching unit, configured to extract feature points with significant geometric structures in each of the images, perform cross-image matching on the feature points in each of the images, and determine feature point pairs with similar geometric structures in each of the images based on the matching results;

[0013] A processing unit, configured to perform three-dimensional reconstruction on the regional points in the measured building area represented by the feature point pairs based on the feature point pairs and the image acquisition parameters to obtain a three-dimensional point cloud of the measured building area, and perform Gaussian distribution processing on the three-dimensional point cloud to obtain a plurality of three-dimensional Gaussian ellipsoids corresponding to the measured building area;

[0014] The image generation unit is used to determine the two-dimensional projection plane of the measured building area and determine the depth information of each three-dimensional Gaussian ellipsoid from the two-dimensional projection plane. Based on the depth information, each three-dimensional Gaussian ellipsoid is splashed into the two-dimensional projection plane to obtain an orthophoto of the measured building area.

[0015] According to a third aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned orthophoto generation method based on three-dimensional Gaussian splattering.

[0016] According to a fourth aspect of the present invention, there is provided a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned orthophoto generation method based on three-dimensional Gaussian splattering when executing the program.

[0017] According to a method, device and storage medium for generating orthophotos based on three-dimensional Gaussian splattering provided by the present invention, compared with the current method of generating orthophotos through digital elevation models, the present invention photographs the measured building from multiple perspectives to obtain multi-view images, and determines feature point pairs with similar geometric structures in each image, and then performs three-dimensional reconstruction of regional points in the measured building area based on the feature point pairs and image acquisition parameters to obtain a three-dimensional point cloud. In the process of generating orthophotos, by constructing a three-dimensional point cloud, more accurate and detailed building information, including information on blocked areas, can be obtained, so that when generating orthophotos, the real form and structural characteristics of the building can be more accurately reflected. The three-dimensional point cloud is processed into multiple three-dimensional Gaussian ellipsoids, and finally each three-dimensional Gaussian ellipsoid is splashed onto the two-dimensional projection plane corresponding to the measured building to obtain the orthophoto. In the process of generating the orthophoto, the three-dimensional Gaussian ellipsoid of the measured building is constructed. Since the three-dimensional Gaussian ellipsoid can accurately represent the three-dimensional form of the building and the three-dimensional Gaussian ellipsoid is editable, the shape of the building can be fine-tuned as needed to adapt to different application scenarios. At the same time, the three-dimensional Gaussian ellipsoid can maintain consistency under different viewing angles, thereby enhancing the three-dimensional sense and authenticity of the orthophoto, so as to improve the generation quality and generation accuracy of the orthophoto. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0019] Figure 1 A flow chart of a method for generating orthophotos based on three-dimensional Gaussian splashing provided by an embodiment of the present invention is shown;

[0020] Figure 2 Another flow chart of a method for generating orthophotos based on three-dimensional Gaussian splashing provided by an embodiment of the present invention is shown;

[0021] Figure 3 A schematic structural diagram of an orthophoto generating device based on three-dimensional Gaussian splashing provided by an embodiment of the present invention is shown;

[0022] Figure 4 A schematic structural diagram of another orthophoto generating device based on three-dimensional Gaussian splashing provided by an embodiment of the present invention is shown;

[0023] Figure 5 A schematic diagram of the physical structure of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0024] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0025] At present, the method of generating orthophotos through digital elevation models mainly records terrain elevation information, but the height and shape information of artificial or natural obstacles such as buildings cannot be accurately expressed, which leads to low accuracy in the generation of orthophotos.

[0026] In order to solve the above problems, the embodiment of the present invention provides a method for generating orthophotos based on three-dimensional Gaussian splashing, such as Figure 1 As shown, the method includes:

[0027] 101. In response to an orthophoto image generation signal of a measured building area, obtain a multi-view image obtained by photographing the measured building area from multiple viewing angles, and determine image acquisition parameters of an image acquisition device corresponding to the multi-view image.

[0028] Among them, the image acquisition device can be a camera or other device; the image acquisition parameters include the resolution, frame rate, exposure time, aperture, focal length, sensor size, gain, contrast, saturation, focal length, principal point coordinates, distortion coefficient, etc. of the image acquisition device, as well as parameters such as the position and posture of the image acquisition device when acquiring each image.

[0029] According to the embodiment of the present invention, an image acquisition device carried by a drone or other device is used to shoot the measured building area at various angles to obtain images at various angles, that is, multi-view images. At the same time, the image acquisition parameters such as the resolution and resolution of the image acquisition device can be obtained by calibrating the image acquisition device.

[0030] 102. Extract feature points with significant geometric structures in each image, perform cross-image matching on the feature points in each image, and determine feature point pairs with similar geometric structures in each image based on the matching results.

[0031] Among them, the feature points with significant geometric structures refer to points with unique properties such as corner points and edge points in the measured building area.

[0032] For the embodiments of the present invention, in order to improve the extraction effect of feature points, it is first necessary to pre-process each image, such as denoising, image enhancement, equalization, etc., so as to improve the quality of each image and facilitate the effective extraction of subsequent feature points. Further, after pre-processing each image, a feature point detection algorithm, such as a scale-invariant feature transformation algorithm and a directional fast detection algorithm, can be used to extract feature points with significant geometric structures such as corner points and edge points in each image, and determine the local area feature vector of each feature point in each image. Then, based on the local area feature vector of each feature point, the similarity between each feature point is calculated across images. For example, based on the local area feature vector corresponding to the feature point a in image A, the Euclidean distance calculation method can be used to calculate the similarity between the feature point a and each feature point in each other image, and the feature point with the greatest similarity in each other image is determined as the similar feature point of the feature point a, and the feature point a and its corresponding similar feature point constitute a feature point pair. For example, the feature point B with the greatest similarity determined in image B b , the feature point C with the maximum similarity determined in image C f , the feature point C with the maximum similarity determined in image D a , by feature point a and feature point B b Constitute a feature point pair, consisting of feature point a and feature point C f Constitute a feature point pair, consisting of feature point a and feature point C a Constitute a pair of feature points. Furthermore, in order to improve the matching accuracy of feature points, two-way matching or nearest neighbor ratio test can also be used to screen matching feature point pairs, and then based on the feature point pairs, construct a three-dimensional point cloud of the measured building area. The embodiment of the present invention matches feature point pairs. Since feature point pairs are points with significant geometric structures in the image, they have a high degree of consistency in images of different perspectives. By determining these feature point pairs, it can be ensured that when constructing a three-dimensional point cloud, images of different perspectives can be accurately matched, which helps to reduce the three-dimensional point cloud construction error caused by factors such as perspective differences and shooting errors, improve the accuracy of determining the three-dimensional point cloud, and further improve the accuracy of generating orthophotos.

[0033] 103. Based on the feature point pairs and the image acquisition parameters, three-dimensionally reconstruct the regional points in the measured building area represented by the feature point pairs to obtain a three-dimensional point cloud of the measured building area, and perform Gaussian distribution processing on the three-dimensional point cloud to obtain multiple three-dimensional Gaussian ellipsoids corresponding to the measured building area.

[0034] For the embodiment of the present invention, the basic matrix between each image can be estimated by using the matched feature point pairs and image acquisition parameters, wherein the basic matrix describes the geometric relationship between the corresponding points in different images. Afterwards, according to the basic matrix and the matched feature point pairs, the position of the regional points of the measured building area in the three-dimensional space can be restored by using the triangulation method, wherein the triangulation is to find the corresponding points in two different images, calculate their positions in the three-dimensional space, and optimize the points in the three-dimensional space as a whole by methods such as bundle adjustment to improve the accuracy and consistency of three-dimensional reconstruction. The coordinates of a series of three-dimensional points in the measured building area can be obtained by the above method, and the three-dimensional points with coordinates constitute the three-dimensional point cloud of the measured building area. Since the feature point pairs are points with significant geometric structures in the image, they have a high degree of consistency in images of different perspectives. Therefore, by constructing the three-dimensional point cloud through the feature point pairs, it can be ensured that the images of different perspectives can be accurately matched, which helps to reduce the three-dimensional point cloud construction error caused by factors such as perspective differences and shooting errors, so that the embodiment of the present invention improves the construction accuracy of the three-dimensional point cloud, and further improves the generation accuracy of the orthophoto image of the measured building area determined based on the three-dimensional point cloud. Furthermore, an orthophoto of the measured building area is generated based on the three-dimensional point cloud. Since the three-dimensional point cloud can describe the building information more comprehensively and in detail, including the information of the obstructed area, the embodiment of the present invention can more accurately reflect the real shape and structural characteristics of the building by constructing a three-dimensional point cloud during the orthophoto generation process, thereby improving the accuracy of orthophoto generation.

[0035] Furthermore, the coordinates, color, opacity and other information of each three-dimensional point in the three-dimensional point cloud are obtained, and at the same time, the covariance matrix of each three-dimensional point is determined, and the three-dimensional Gaussian ellipsoid corresponding to each three-dimensional point is determined according to the coordinates, color, opacity and covariance matrix of each three-dimensional point, and finally the orthophoto of the measured building area is determined according to the three-dimensional Gaussian ellipsoid. The embodiment of the present invention generates an orthophoto by constructing a three-dimensional Gaussian ellipsoid. Since the three-dimensional Gaussian ellipsoid can accurately represent the three-dimensional form of the building, and the three-dimensional Gaussian ellipsoid is editable, the shape of the building can be fine-tuned as needed to adapt to different application scenarios. At the same time, the three-dimensional Gaussian ellipsoid can maintain consistency under different viewing angles, thereby enhancing the stereoscopic sense and authenticity of the orthophoto, so as to improve the generation quality and generation accuracy of the orthophoto.

[0036] 104. Determine the two-dimensional projection plane of the measured building area, and determine the depth information of each three-dimensional Gaussian ellipsoid from the two-dimensional projection plane. Based on the depth information, splash each three-dimensional Gaussian ellipsoid into the two-dimensional projection plane to obtain an orthophoto of the measured building area.

[0037] For the embodiment of the present invention, the two-dimensional projection plane of the measured building area is determined in advance according to actual needs, and then the difference between the Z coordinate of each point on the three-dimensional Gaussian ellipsoid (assuming that the two-dimensional projection plane is the XY plane) and the Z coordinate of the corresponding point on the two-dimensional projection plane is calculated to obtain depth information. According to the calculated depth information, each point on each three-dimensional Gaussian ellipsoid can be splashed according to its vertical distance to the two-dimensional projection plane, so as to obtain an orthophoto of the measured building area. In the process of generating an orthophoto, the present invention constructs a three-dimensional Gaussian ellipsoid of the measured building. Since the three-dimensional Gaussian ellipsoid can accurately represent the three-dimensional form of the building, and the three-dimensional Gaussian ellipsoid is editable, the shape of the building can be fine-tuned as needed to adapt to different application scenarios. At the same time, the three-dimensional Gaussian ellipsoid can maintain consistency under different viewing angles, thereby enhancing the stereoscopic sense and authenticity of the orthophoto, so as to improve the generation quality and generation accuracy of the orthophoto.

[0038] According to a method for generating an orthophoto based on three-dimensional Gaussian splattering provided by the present invention, compared with the current method of generating orthophotos through digital elevation models, the present invention photographs the measured building from multiple perspectives to obtain multi-view images, and determines feature point pairs with similar geometric structures in each image, and then performs three-dimensional reconstruction of regional points in the measured building area based on the feature point pairs and image acquisition parameters to obtain a three-dimensional point cloud. In the process of orthophoto generation, by constructing a three-dimensional point cloud, more accurate and detailed building information, including occlusion, can be obtained, so that when generating orthophotos, the real form and structural characteristics of the building can be more accurately reflected, thereby improving the orthophoto quality. The accuracy of generating the orthophoto is improved, and then the three-dimensional point cloud is processed into multiple three-dimensional Gaussian ellipsoids, and finally each three-dimensional Gaussian ellipsoid is splashed onto the two-dimensional projection plane corresponding to the measured building to obtain the orthophoto. In the process of generating the orthophoto, the present invention constructs a three-dimensional Gaussian ellipsoid of the measured building. Since the three-dimensional Gaussian ellipsoid can finely represent the three-dimensional form of the building and the three-dimensional Gaussian ellipsoid is editable, the shape of the building can be fine-tuned as needed to adapt to different application scenarios. At the same time, the three-dimensional Gaussian ellipsoid can maintain consistency under different viewing angles, thereby enhancing the three-dimensional sense and authenticity of the orthophoto, so as to improve the generation quality and generation accuracy of the orthophoto.

[0039] Further, in order to better illustrate the above process of generating orthophotos, as a refinement and extension of the above embodiment, the embodiment of the present invention provides another method for generating orthophotos based on three-dimensional Gaussian splattering, such as Figure 2 As shown, the method includes:

[0040] 201. In response to an orthophoto image generation signal of a measured building area, obtain a multi-view image obtained by photographing the measured building area from multiple viewing angles, and determine image acquisition parameters of an image acquisition device corresponding to the multi-view image.

[0041] Specifically, a camera or other device is used to shoot the building area under test from various angles to obtain multi-view images, and the camera is calibrated to obtain image acquisition parameters of the image acquisition device.

[0042] 202. Extract feature points with significant geometric structures in each image, perform cross-image matching on the feature points in each image, and determine feature point pairs with similar geometric structures in each image based on the matching results.

[0043] For the embodiment of the present invention, in order to improve the accuracy of orthophoto generation, it is first necessary to preprocess the multi-view images, that is, the images obtained by shooting the measured building area from multiple angles. Based on this, the method includes: taking each pixel point in each of the images as a pixel point to be processed, determining the pixel mean of each pixel point in a preset neighborhood corresponding to each pixel point to be processed in each of the images; using the pixel mean to replace the pixel value of the corresponding pixel point to be processed, to obtain each of the preprocessed images.

[0044] Among them, the preset area is set according to actual needs. Specifically, for example, for image A, the pixels contained in image A are pixel a, pixel b, and pixel c. The pixel mean A of each pixel in the preset neighborhood of pixel a is determined, and the pixel value of pixel a is replaced by the pixel mean A. The pixel mean B of each pixel in the preset neighborhood of pixel b is determined, and the pixel value of pixel b is replaced by the pixel mean B. The pixel mean C of each pixel in the preset neighborhood of pixel c is determined, and the pixel value of pixel c is replaced by the pixel mean C. After all the pixels in image A are replaced, the preprocessing of image A is completed, and thus each image can be preprocessed in the above manner to obtain each preprocessed image.

[0045] Further, feature points with significant geometric structures are extracted from each preprocessed image. Based on this, step 202 specifically includes: taking each of the preprocessed images as the bottommost pyramid layer image of the image pyramid to be constructed; taking each of the pyramid layers except the bottommost pyramid layer in the image pyramid to be constructed as a target pyramid layer; based on the number of rows and columns of the image in the previous pyramid layer corresponding to the target pyramid layer, sequentially performing Gaussian blur processing and downsampling processing on the image in the previous pyramid layer to obtain the target pyramid layer image; and constructing an image pyramid corresponding to each of the images based on the bottommost pyramid layer image and each of the target pyramid layer images; processing the images in two adjacent pyramid layers in the image pyramid into images with the same resolution, and subtracting the images with the same resolution in the two adjacent pyramid layers to obtain a Gaussian difference pyramid. The method for processing the images in two adjacent pyramid layers in the image pyramid into images with the same resolution includes: determining the low-resolution image layer and the high-resolution image layer in the two adjacent pyramid layers. The method comprises the steps of: determining a magnification coefficient of an image in the low-resolution image layer based on the number of rows and columns of the image in the high-resolution image layer and the number of rows and columns of the image in the low-resolution image layer; performing upsampling processing on the image in the low-resolution image layer based on the magnification coefficient to obtain an image with the same resolution that matches the resolution of the image in the high-resolution image layer; using the difference pyramid layers except the first layer and the last layer in the Gaussian difference pyramid as target difference pyramid layers, and using each pixel of the image in the target difference pyramid layer as a target pixel; determining each adjacent evaluation pixel in a preset difference neighborhood corresponding to the target pixel in the difference pyramid layer adjacent to the target difference pyramid layer, and determining each same-layer evaluation pixel in a preset same-layer neighborhood corresponding to the target pixel in the target difference pyramid layer; if the target pixel is the maximum pixel value point or the minimum pixel value point among the adjacent evaluation pixels and the same-layer evaluation pixels, then determining the target pixel as a feature point with a significant geometric structure in the corresponding image.

[0046] Among them, the preset differential neighborhood and the preset same-layer neighborhood are set according to actual needs. Specifically, in order to extract the feature points in each preprocessed image, it is first necessary to construct an influence pyramid for each image. The specific construction method of the image pyramid is: take any image A in each image as an example, first use image A as the bottom-level image of the image pyramid. This layer of image has the highest resolution and is the clearest. Then, Gaussian blur processing is performed on the bottom-level image. After Gaussian blur processing, downsampling processing is performed. The downsampling process is such as merging every 2×2 pixels in the bottom-level image into a new pixel, and all new pixels constitute the image in the upper pyramid layer corresponding to the bottom level. If the upper pyramid layer corresponding to the bottom level is determined as the second pyramid layer, the image in the second pyramid layer continues to be processed. Gaussian blur processing, after Gaussian blur processing, downsampling processing is performed. The downsampling process is such as merging every 2×2 pixels in the image in the second pyramid layer into a new pixel, and all new pixels constitute the image in the previous pyramid layer corresponding to the second pyramid layer, that is, the image in the third pyramid layer, and so on. Gaussian blur processing and downsampling processing are performed on the image in the previous pyramid layer layer by layer to obtain the image in the current pyramid layer, until the resolution of the image in the latest pyramid layer meets the requirements, and finally the image pyramid of image A is composed of various pyramid layers. Similarly, the image pyramid of each image can be constructed in the above manner, and the resolution of the image in each pyramid layer in the image pyramid is different.

[0047] Furthermore, after determining that the image pyramid corresponding to each image is constructed, it is also necessary to construct a Gaussian difference pyramid. The specific construction scheme is: taking the two adjacent pyramid layers of the first pyramid layer and the second pyramid layer in the image pyramid as an example, since the resolutions of the images in the first pyramid layer and the second pyramid layer are different, the high-resolution image layer and the low-resolution image layer are first determined in the first pyramid layer and the second pyramid layer, and the resolutions of the images in the high-resolution image layer and the low-resolution image layer are respectively determined, that is, the number of rows and columns of the images, based on the resolutions of the images in the high-resolution image layer and the low-resolution image layer, the magnification factor of the image in the low-resolution image layer is determined, and the image in the low-resolution image layer is upsampled according to the magnification factor. The specific upsampling method is such as bilinear interpolation, bicubic interpolation, nearest pixel method, etc., so that the resolution of the image in the low-resolution image layer is processed to be the same as the resolution of the image in the high-resolution image layer. Therefore, according to the above method, the images in each adjacent pyramid layer in the image pyramid can be processed as images with the same resolution, and then Subtract the images of the same resolution in two adjacent pyramid layers in the image pyramid to obtain a Gaussian difference pyramid. The Gaussian difference pyramid is recorded from the bottom layer to the top layer in sequence as the first difference pyramid layer, the second difference pyramid layer, the third difference pyramid layer, etc. Starting from the second difference pyramid layer, the adjacent evaluation pixels and the same-layer evaluation pixels of each pixel in the second difference pyramid layer are determined. The specific determination method takes the pixel a in the second difference pyramid layer as an example. In the first difference pyramid layer and the third difference pyramid layer, each pixel in the preset difference neighborhood corresponding to the pixel a is determined, that is, the adjacent evaluation pixels. At the same time, in the second difference pyramid layer, each pixel in the preset same-layer neighborhood corresponding to the pixel a is determined, that is, the same-layer evaluation pixels. Then, it is determined whether the pixel a is the pixel with the largest pixel value or the smallest pixel value among each adjacent evaluation pixel and each same-layer evaluation pixel. If so, the pixel a is determined as a feature point in the corresponding image, thereby being able to determine feature points with significant geometric structures in each image. When the embodiment of the present invention uses the Gaussian difference pyramid to extract feature points, since image information at different scales is taken into account, the extracted feature points are scale invariant, that is, the same feature points can be accurately identified at different scales, thereby improving the accuracy of extracting feature points in the image.

[0048] Furthermore, after extracting the feature points in each image, the feature points in each image are matched across images for similarity, so as to determine the feature point pairs with similar geometric structures in each image, and then the mismatched point pairs can be filtered through geometric constraints to ensure that only geometrically reasonable feature point pairs are retained. The relative positions between images can also be further optimized to improve the geometric consistency and the stability of feature point pair matching.

[0049] 203. Based on the feature point pairs and the image acquisition parameters, three-dimensionally reconstruct the regional points in the measured building area represented by the feature point pairs to obtain a three-dimensional point cloud of the measured building area.

[0050] After determining the feature point pairs in each image, in order to improve the generation accuracy of the orthophoto image, it is also necessary to determine the three-dimensional point cloud of the measured building area. Based on this, step 203 specifically includes: based on the feature point pairs, generating an image association graph corresponding to each of the images, wherein the nodes in the image association graph represent each of the images, and the edges in the image association graph represent the geometric relationship between the two connected images, and the geometric relationship refers to the relative position and perspective difference between the two images obtained through the feature point pairs; based on the geometric relationship between the two connected images in the image association graph, determining the basic matrix and the essential matrix between the two connected images; based on the basic matrix and the essential matrix, determining the relative posture between the two connected images, and based on the relative posture between each of the two connected images, determining the global image posture corresponding to each of the images, wherein the relative posture is the relative rotation matrix and translation vector between the two connected images; based on the global image posture and the image acquisition parameters, performing three-dimensional reconstruction on the regional points in the measured building area represented by the feature point pairs to obtain the three-dimensional point cloud of the measured building area.

[0051] Specifically, according to the feature matching results between images, that is, the feature point pairs, an image association graph of the images is generated. The nodes of the image association graph represent images, and the edges of the image association graph represent the geometric relationships between images obtained by matching the feature points, reflecting the relative positions and perspective differences of the images. Then, based on the geometric relationship between the two connected images in the image association graph, the basic matrix between the two connected images is determined. The specific method for determining the basic matrix is: taking any two adjacent images in the image association graph as an example, and all the feature point pairs between the two adjacent images, a feature point pair set is constructed, and then the eight-point method can be used to calculate the basic matrix, such as in Determine at least 8 pairs of matching points in the feature point pair set to calculate the basic matrix, set the unknown elements of the basic matrix to be constructed as variables, and construct a linear equation system containing these variables, and then use 8 pairs of matching points to solve the linear equation system. After solving the linear equation system, a candidate solution of the basic matrix will be obtained. In order to obtain a unique basic matrix that meets actual needs, it is necessary to perform singular value decomposition on the candidate solution. Through singular value decomposition, the basic matrix can be decomposed into the product of three matrices, and the decomposed matrix is ​​adjusted to meet the property that the rank of the basic matrix is ​​2, so as to obtain the basic matrix between two adjacent images. Furthermore, it is also necessary to determine the essential matrix between two adjacent images. The specific determination method is to first determine the intrinsic parameter matrix of the image acquisition device, and determine the essential matrix between the two connected images through the intrinsic parameter matrix and the basic matrix, wherein the essential matrix describes the relative posture relationship between the two connected images.

[0052] Further, after determining the basic matrix and the essential matrix between the two connected images, singular value decomposition is performed on the basic matrix and the essential matrix to obtain an orthogonal matrix and a diagonal matrix. The rotation matrix and the translation vector are determined according to the orthogonal matrix and the diagonal matrix. The rotation matrix and the translation vector are used to represent the relative position and orientation between the two connected images, so as to obtain the relative posture between the two connected images. In an image association graph, if the relative posture between two adjacent images is known, the global image posture of each image in the entire image association graph can be estimated by accumulating these relative postures. Further, after determining the global image posture corresponding to each image, for each pair of matching feature points, their two-dimensional coordinates on different images, global image posture and image acquisition parameters (image acquisition parameters such as the intrinsic parameters of the camera: focal length, principal point coordinates, radial distortion coefficient, etc.) are used to calculate their three-dimensional coordinates through triangulation, and all three-dimensional points with three-dimensional coordinates are integrated together to obtain a three-dimensional point cloud of the measured building area. The embodiment of the present invention constructs a three-dimensional point cloud of the measured building area by matching feature point pairs. Since the special points in the feature point pairs are usually significant structures in the image, their positions in the three-dimensional space are relatively independent and stable. Therefore, when constructing the three-dimensional point cloud, only these feature points can be used to generate point cloud data, thereby avoiding the generation of a large amount of redundant data, helping to reduce the amount of data for subsequent processing, improving processing efficiency, and further improving the efficiency of generating orthophotos.

[0053] Furthermore, after determining the three-dimensional point cloud of the measured building area, in order to more accurately represent the shape and details of the measured building area, it is necessary to pre-construct the three-dimensional point cloud into a dense three-dimensional point cloud, that is, to increase the number of point clouds. Based on this, the method includes: judging whether the point cloud density of the three-dimensional point cloud is less than a preset density threshold; if it is greater than or equal to the preset density threshold, determining that the three-dimensional point cloud is a dense three-dimensional point cloud; if it is less than the preset density threshold, determining each reference point in a preset local neighborhood corresponding to each three-dimensional point in the three-dimensional point cloud; based on each of the reference points, using a preset weighted least squares method to select the reference points in the corresponding three-dimensional point cloud. A local surface should be fitted within a preset local neighborhood, wherein the geometric similarity between the local surface and the actual surface formed by the reference points within the corresponding preset local neighborhood meets the preset requirements; the number of three-dimensional points in the three-dimensional point cloud is determined, and based on the number of three-dimensional points, multiple groups of external points are randomly generated in the external three-dimensional space of the three-dimensional point cloud; each group of external points is projected onto the local surface corresponding to the corresponding three-dimensional point to obtain the local surface projection point corresponding to each three-dimensional point, and each local surface projection point is determined as a newly added three-dimensional point, and a dense three-dimensional point cloud is formed by the three-dimensional point cloud and the newly added three-dimensional points.

[0054] Among them, the preset density threshold is set according to actual needs, and the preset local neighborhood is set according to actual needs. Specifically, if the point cloud density of the three-dimensional point cloud is greater than or equal to the preset density threshold, the three-dimensional point cloud can be directly defined as a dense three-dimensional point cloud. At this time, the dense three-dimensional point cloud is directly subjected to Gaussian distribution processing to obtain multiple three-dimensional Gaussian ellipsoids corresponding to the measured building area. If the point cloud density of the three-dimensional point cloud is less than the preset density threshold, the three-dimensional point cloud is defined as a sparse three-dimensional point cloud. For the sparse three-dimensional point cloud, it is necessary to add three-dimensional points to construct a dense three-dimensional point cloud. The specific method of constructing a dense three-dimensional point cloud is as follows: taking any three-dimensional point a in the three-dimensional point cloud as an example, determine each three-dimensional point, that is, each reference point, within the preset local neighborhood of the three-dimensional point a, and then construct a mathematical model that can fit the local surface according to the principle of weighted least squares method. , the mathematical model is usually a polynomial surface or other types of surface models, whose parameters need to be determined through the fitting process, and the initial parameters are set for the fitting model. These parameters can be random or based on some prior knowledge. After that, the parameters of the fitting model are continuously adjusted through iterative calculation to minimize the weighted square error. In each iteration, the fitting surface is calculated according to the current parameters, and the weighted square error from each reference point to the fitting surface is calculated. The optimization algorithm is used to update the parameters to reduce the error. The condition for the convergence of the iterative process can be that the error reaches a preset threshold, or the number of iterations reaches a preset upper limit. When the convergence condition is met, the iteration is stopped and the final fitting surface is output, that is, the local surface corresponding to the three-dimensional point a. The similarity between the local surface and the actual surface composed of each three-dimensional point in the preset local neighborhood corresponding to the three-dimensional point a is greater than the preset threshold (the preset threshold is set according to actual needs). Further, according to the number of points in the three-dimensional point cloud, multiple groups of external points are randomly generated in the external three-dimensional control of the three-dimensional point cloud, such as a group of external points corresponding to the three-dimensional point a, and a group of external points corresponding to the three-dimensional point b are projected on the local surface corresponding to the three-dimensional point a, and each projection point on the local surface is obtained. Thus, according to the above method, each projection point on the local surface corresponding to each three-dimensional point in the three-dimensional point cloud can be determined, and all projection points corresponding to each three-dimensional point are determined as newly added three-dimensional points, and then a dense three-dimensional point cloud is formed by the newly added three-dimensional points and all projection points, and finally the three-dimensional Gaussian ellipsoid of the measured building is determined according to the dense three-dimensional point cloud. In the case where the number of three-dimensional points in the three-dimensional point cloud is small, the embodiment of the present invention adjusts the point cloud density by increasing the number of three-dimensional points, so as to ensure that the geometric details and visual effects in the scene of the measured building area can better reflect reality, and increase the Gaussian points to enhance the performance of the details. In addition, in the case where the number of three-dimensional points in the three-dimensional point cloud is large, the embodiment of the present invention can also delete redundant points to reduce errors, and modify the position and shape of the points to adapt to more accurate geometric shapes.

[0055] 204. Determine the three-dimensional coordinates of each three-dimensional point in the three-dimensional point cloud, and determine the three-dimensional point color and the three-dimensional point opacity of each three-dimensional point.

[0056] Specifically, before constructing the three-dimensional point cloud, the three-dimensional coordinates of each three-dimensional point are determined, and the color information and opacity information of each three-dimensional point can be obtained through the original image where the matched feature point is located.

[0057] 205. Determine the point cloud center of the three-dimensional point cloud, and determine the covariance matrix corresponding to each three-dimensional point based on the distance and direction from each three-dimensional point to the point cloud center.

[0058] Among them, the covariance matrix is ​​a matrix that describes the linear relationship between each dimension in the point cloud data. Specifically, the center of the point cloud can be determined by the centroid or center point of the three-dimensional point cloud. The coordinates of each point in the three-dimensional point cloud are subtracted from the coordinates of the center point of the point cloud to obtain the three-dimensional point cloud after the centroid is removed. Based on the coordinate vector of each point in the three-dimensional point cloud after the centroid is removed and the average coordinate vector of the three-dimensional point cloud after the centroid is removed, the covariance matrix corresponding to each three-dimensional point is constructed, wherein the diagonal elements of the covariance matrix represent the variance of each three-dimensional point in each dimension (i.e., the average value of the square of the deviation of the X, Y, and Z coordinates), and the non-diagonal elements represent the covariance of each three-dimensional point between different dimensions (i.e., the measure of the linear relationship between X and Y, X and Z, and Y and Z).

[0059] 206. Perform eigenvalue decomposition on the covariance matrix to obtain a rotation matrix and a scaling matrix corresponding to each three-dimensional point.

[0060] Specifically, after determining the covariance matrix of each three-dimensional point, the covariance matrix is ​​decomposed into eigenvalue matrix and eigenvector matrix, where the eigenvalue represents the variance in each main direction, and the eigenvector represents these directions. The rotation matrix can be directly constructed from the eigenvector matrix of the covariance matrix, that is, the column vectors of the eigenvector matrix are used as the column vectors of the rotation matrix to obtain an orthogonal matrix (i.e., the rotation matrix). The scaling matrix can be constructed through the eigenvalue matrix. The scaling matrix is ​​a diagonal matrix, and the elements on its diagonal are the square roots of the eigenvalues ​​of the covariance matrix.

[0061] 207. Determine the three-dimensional Gaussian ellipsoid corresponding to each three-dimensional point based on the three-dimensional coordinates, three-dimensional point colors, three-dimensional point opacity, covariance matrix, rotation matrix, and scaling matrix corresponding to each three-dimensional point.

[0062] Specifically, the ellipsoid center of the three-dimensional Gaussian ellipsoid is the three-dimensional coordinates of the corresponding three-dimensional point, the covariance matrix of the three-dimensional point determines the shape of the corresponding three-dimensional Gaussian ellipsoid, the rotation matrix of the three-dimensional point determines the direction of the corresponding three-dimensional Gaussian ellipsoid, the scaling matrix of the three-dimensional point determines the size of the corresponding three-dimensional Gaussian ellipsoid, the three-dimensional point color of the three-dimensional point determines the ellipsoid color of the corresponding three-dimensional Gaussian ellipsoid, and the three-dimensional point opacity of the three-dimensional point determines the opacity of the corresponding three-dimensional Gaussian ellipsoid.

[0063] Furthermore, in order to improve the reconstruction accuracy of the measured building area and fit the three-dimensional shape of the building more accurately, after constructing the three-dimensional Gaussian ellipsoid of the measured building area, it is also necessary to optimize the three-dimensional Gaussian ellipsoid. Based on this, the method includes: constructing a parameter optimization bounding box of the three-dimensional Gaussian ellipsoid based on the position of each three-dimensional point in the three-dimensional point cloud in the three-dimensional coordinate system; defining each of the images as a reference image, and based on the shooting angle corresponding to each of the reference images, projecting each of the three-dimensional Gaussian ellipsoids to the projection plane corresponding to each of the shooting angles, to obtain a projection image corresponding to each of the reference images; comparing the projection images with the corresponding reference images for similarity, and based on the comparison results, determining the parameter optimization function corresponding to each of the three-dimensional Gaussian ellipsoids; based on the parameter optimization function, optimizing the parameters of the three-dimensional Gaussian ellipsoid within the parameter optimization bounding box to obtain the three-dimensional Gaussian ellipsoid after optimizing the parameters. Gaussian ellipsoid; wherein the parameters include the direction, size, color, and opacity of the three-dimensional Gaussian ellipsoid; each of the images is input into a large model for depth estimation to obtain a predicted depth information map of the measured building area; a top-view projection plane of the measured building area is determined, and the three-dimensional Gaussian ellipsoid after each of the optimized parameters is projected onto the top-view projection plane to obtain a top-view projection image; a depth value of each pixel in the top-view projection image is determined, and based on the depth value of each pixel, a top-view depth information map of the measured building area is determined; based on the predicted depth information map and the top-view depth information map, a position loss function of each three-dimensional Gaussian ellipsoid after the optimized parameters is determined, and based on the position loss function, the ellipsoid center of each three-dimensional Gaussian ellipsoid after the optimized parameters is optimized in the parameter optimization bounding box, and the three-dimensional Gaussian ellipsoid is replaced by the three-dimensional Gaussian ellipsoid after the position is optimized.

[0064] Specifically, according to the positions of different camera perspectives in the world coordinate system and the positions of the three-dimensional points, a parameter optimization bounding box containing all perspectives is selected as the optimization range to ensure that the parameter optimization does not exceed the optimization range. First, according to the shooting perspective of each image, the projection plane corresponding to each image is determined. For example, when the camera is shooting directly at the building, the projection plane is usually parallel to the front of the building. When the camera shoots the building from the side, the projection plane is parallel to the side of the building. Then, each three-dimensional Gaussian ellipsoid is projected on the corresponding projection plane to obtain the projection image corresponding to each image. Then, the image feature vector corresponding to each image and the projection image feature vector corresponding to each projection image are determined. , and based on the image feature vector and its corresponding projection image feature vector, calculate the similarity between each image and its corresponding projection image. If the similarity is greater than a preset similarity threshold (the preset similarity threshold is set according to actual needs), there is no need to optimize the parameters of the three-dimensional Gaussian ellipsoid corresponding to the image. If the similarity is less than or equal to the preset similarity threshold, it is necessary to determine the parameter optimization function of the three-dimensional Gaussian ellipsoid corresponding to the image according to the gap between the image and its corresponding projection image. Finally, use the parameter optimization function to optimize the direction, size, color, opacity and other parameters of the three-dimensional Gaussian ellipsoid corresponding to the image to obtain the three-dimensional Gaussian ellipsoid after optimizing the parameters.

[0065] Furthermore, the above parameter optimization method can optimize the direction, size, color, and opacity of the three-dimensional Gaussian ellipsoid. In addition, the ellipsoid center of the three-dimensional Gaussian ellipsoid after the optimized parameters needs to be optimized. The specific optimization method is: first, use the large model to predict the predicted depth information map corresponding to the measured building area, where the depth map is a grayscale image, in which each pixel carries the depth information about the point from the camera or sensor. This information is usually expressed in grayscale values, where brighter pixels represent objects closer to the camera, and darker pixels represent objects farther from the camera. At the same time, the parameterized three-dimensional Gaussian ellipsoid is projected onto the overhead projection plane of the measured building area to obtain the overhead projection image, and the depth information of each pixel point in the overhead projection image from the camera or sensor is determined, thereby obtaining the overhead depth information map of the measured building area, and then the depth residual at the corresponding pixel position in the overhead depth information map and the predicted depth information map is determined. Based on the depth residual, the geometric position loss of the Gaussian ellipsoid in space is defined, that is, the position loss function. The position loss function can be a mean square error or a more complex weighted loss function. Then, the position of the center of the three-dimensional Gaussian ellipsoid is iteratively optimized in the parameter optimization bounding box using algorithms such as the gradient descent method. In each iteration, the loss value under the current parameters is calculated according to the position loss function, and the position of the center of the ellipsoid is updated according to the optimization algorithm. The iterative process is repeated until the convergence conditions are met, such as the loss value reaches a preset threshold, the number of iterations reaches an upper limit, and other convergence conditions. When the optimization process ends, the optimized ellipsoid center position is output, and finally the three-dimensional Gaussian ellipsoid is replaced by the three-dimensional Gaussian ellipsoid with the optimized position. The embodiment of the present invention optimizes parameters such as color, position, opacity, etc. in the three-dimensional Gaussian ellipsoid, thereby obtaining a better geometric representation of the measured building area, reducing the geometric distortion in the orthophoto, and refining the representation of the scene, providing a realistic and accurate rendering effect for generating orthophotos, thereby improving the generation accuracy of orthophotos.

[0066] 208. Determine the two-dimensional projection plane of the measured building area, and determine the depth information of each three-dimensional Gaussian ellipsoid from the two-dimensional projection plane. Based on the depth information, splash each three-dimensional Gaussian ellipsoid into the two-dimensional projection plane to obtain an orthophoto of the measured building area.

[0067] For the embodiment of the present invention, after determining multiple three-dimensional Gaussian ellipsoids corresponding to the measured building area, it is necessary to generate an orthophoto of the measured building area based on the three-dimensional Gaussian ellipsoids. Based on this, step 208 specifically includes: based on the depth information, sorting each of the three-dimensional Gaussian ellipsoids to obtain the sorted three-dimensional Gaussian ellipsoids, and based on the depth information, determining the pixel projection weight of each of the three-dimensional Gaussian ellipsoids; using a rasterization algorithm to project the surface points on each of the sorted three-dimensional Gaussian ellipsoids to the two-dimensional projection plane according to position and pixel, and based on the pixel projection weight, superimposing the surface points on each of the three-dimensional Gaussian ellipsoids onto the pixel values ​​on the two-dimensional projection plane to obtain an orthophoto of the measured building area.

[0068] Specifically, the two-dimensional projection plane of the measured building area is determined according to actual needs, and based on each point on the three-dimensional Gaussian ellipsoid, the vertical distance from each point to the two-dimensional projection plane is calculated, and the shortest vertical distance is determined as the depth information of the corresponding three-dimensional Gaussian ellipsoid, and then the three-dimensional Gaussian ellipsoids are sorted in order from small to large according to the depth information to obtain the sorted three-dimensional Gaussian ellipsoids, and then the sorted three-dimensional Gaussian ellipsoids are splashed onto the two-dimensional projection plane in sequence according to the snowball throwing method to obtain the projection point of each three-dimensional Gaussian ellipsoid on the two-dimensional projection plane, and the position of the coordinates of the projection point on the two-dimensional projection plane is calculated, and according to the position of the projection point, the shape and size of the ellipsoid, and the depth information , calculate the weight of each projection point on the corresponding pixel on the two-dimensional projection plane, that is, the pixel projection weight, and superimpose the attributes of each projection point (such as color, brightness or depth information) on the value of the corresponding pixel on the two-dimensional projection plane according to its pixel projection weight to obtain an orthophoto. If multiple projection points are projected onto the same pixel, they are weighted averaged or the maximum value is selected according to their weights. When superimposing pixel values, edge and occlusion problems are considered. Depth testing or occlusion culling algorithms can be used to deal with these problems. The generated projection image is post-processed, such as smoothing, contrast adjustment, color correction, etc., to obtain a corrected orthophoto of the measured building area.

[0069] Another orthophoto generation method based on three-dimensional Gaussian splashing provided by the present invention, compared with the current method of generating orthophotos through digital elevation models, the present invention photographs the measured building from multiple perspectives to obtain multi-view images, and determines feature point pairs with similar geometric structures in each image, and then performs three-dimensional reconstruction of regional points in the measured building area based on the feature point pairs and image acquisition parameters to obtain a three-dimensional point cloud. In the process of orthophoto generation, by constructing a three-dimensional point cloud, more accurate and detailed building information, including occlusion, can be obtained, so that when generating orthophotos, the real form and structural characteristics of the building can be more accurately reflected, thereby improving The accuracy of orthophoto generation is improved, and then the three-dimensional point cloud is processed into multiple three-dimensional Gaussian ellipsoids, and finally each three-dimensional Gaussian ellipsoid is splashed onto the two-dimensional projection plane corresponding to the measured building to obtain the orthophoto. In the process of generating the orthophoto, the present invention constructs a three-dimensional Gaussian ellipsoid of the measured building. Since the three-dimensional Gaussian ellipsoid can finely represent the three-dimensional form of the building, and the three-dimensional Gaussian ellipsoid is editable, the shape of the building can be fine-tuned as needed to adapt to different application scenarios. At the same time, the three-dimensional Gaussian ellipsoid can maintain consistency under different viewing angles, thereby enhancing the three-dimensional sense and authenticity of the orthophoto, so as to improve the generation quality and generation accuracy of the orthophoto.

[0070] Further, as Figure 1 In a specific implementation, an embodiment of the present invention provides an orthophoto generation device based on three-dimensional Gaussian splashing, such as Figure 3 As shown, the device includes: an acquisition unit 31, a matching unit 32, a processing unit 33, and an image generation unit 34.

[0071] The acquisition unit 31 may be configured to generate a signal in response to an orthophoto of the measured building area, acquire a multi-view image obtained by photographing the measured building area from multiple viewing angles, and determine image acquisition parameters of an image acquisition device corresponding to the multi-view image.

[0072] The matching unit 32 may be used to extract feature points with significant geometric structures in each of the images, perform cross-image matching on the feature points in each of the images, and determine feature point pairs with similar geometric structures in each of the images based on the matching results.

[0073] The processing unit 33 can be used to perform three-dimensional reconstruction on the regional points in the measured building area represented by the feature point pairs based on the feature point pairs and the image acquisition parameters to obtain a three-dimensional point cloud of the measured building area, and perform Gaussian distribution processing on the three-dimensional point cloud to obtain multiple three-dimensional Gaussian ellipsoids corresponding to the measured building area.

[0074] The image generation unit 34 can be used to determine the two-dimensional projection plane of the measured building area, and determine the depth information of each three-dimensional Gaussian ellipsoid from the two-dimensional projection plane. Based on the depth information, each three-dimensional Gaussian ellipsoid is splashed into the two-dimensional projection plane to obtain an orthophoto of the measured building area.

[0075] In specific application scenarios, in order to extract feature points with significant geometric structures in each image, such as Figure 4 As shown, the matching unit 32 includes a first determination module 321 , a replacement module 322 , a construction module 323 , and a subtraction module 324 .

[0076] The first determination module 321 may be configured to use each pixel point in each of the images as a pixel point to be processed, and determine a pixel mean value of each pixel point in a preset neighborhood corresponding to each pixel point to be processed in each of the images.

[0077] The replacement module 322 may be used to replace the pixel value of the corresponding pixel point to be processed with the pixel mean value to obtain the pre-processed images.

[0078] The first determining module 321 may also be configured to use each of the pre-processed images as a bottommost pyramid layer image of the corresponding image pyramid to be constructed.

[0079] The construction module 323 can be used to use each pyramid layer except the bottom pyramid layer in the image pyramid to be constructed as a target pyramid layer, and based on the number of rows and columns of the image in the previous pyramid layer corresponding to the target pyramid layer, sequentially perform Gaussian blur processing and downsampling processing on the image in the previous pyramid layer to obtain the target pyramid layer image, and construct the image pyramid corresponding to each of the images based on the bottom pyramid layer image and the target pyramid layer images.

[0080] The subtraction module 324 can be used to process the images in two adjacent pyramid layers in the image pyramid into images with the same resolution, and subtract the images with the same resolution in the two adjacent pyramid layers to obtain a Gaussian difference pyramid, wherein the method for processing the images in two adjacent pyramid layers in the image pyramid into images with the same resolution includes: determining a low-resolution image layer and a high-resolution image layer in the two adjacent pyramid layers, and determining a magnification factor of the image in the low-resolution image layer based on the number of rows and columns of the image in the high-resolution image layer and the number of rows and columns of the image in the low-resolution image layer; and up-sampling the image in the low-resolution image layer based on the magnification factor to obtain an image with the same resolution that matches the resolution of the image in the high-resolution image layer.

[0081] The first determining module 321 may be specifically configured to use the difference pyramid layers except the first layer and the last layer in the Gaussian difference pyramid as target difference pyramid layers, and use each pixel point of the image in the target difference pyramid layer as a target pixel point.

[0082] The first determining module 321 may be specifically configured to determine, in a differential pyramid layer adjacent to the target differential pyramid layer, each adjacent evaluation pixel within a preset differential neighborhood corresponding to the target pixel, and to determine, in the target differential pyramid layer, each same-layer evaluation pixel within a preset same-layer neighborhood corresponding to the target pixel.

[0083] The first determination module 321 can be specifically used to determine the target pixel point as a feature point with a significant geometric structure in the corresponding image if the target pixel point is the maximum pixel value point or the minimum pixel value point among the adjacent evaluation pixel points and the same-layer evaluation pixel points.

[0084] In a specific application scenario, in order to determine the three-dimensional point cloud of the measured building area, the processing unit 33 includes a generation module 331 , a second determination module 332 , and a three-dimensional reconstruction module 333 .

[0085] The generation module 331 can be used to generate an image association graph corresponding to each of the images based on the feature point pairs, wherein the nodes in the image association graph represent each of the images, and the edges in the image association graph represent the geometric relationship between two connected images, and the geometric relationship refers to the relative position and perspective difference between the two images obtained through the feature point pairs.

[0086] The second determination module 332 may be configured to determine a basic matrix and an essential matrix between two connected images based on a geometric relationship between the two connected images in the image association graph.

[0087] The second determination module 332 can be specifically used to determine the relative posture between the two connected images based on the basic matrix and the essential matrix, and determine the global image posture corresponding to each image based on the relative posture between each of the two connected images, wherein the relative posture is the relative rotation matrix and translation vector between the two connected images.

[0088] The three-dimensional reconstruction module 333 can be used to perform three-dimensional reconstruction on the regional points in the measured building area represented by the feature point pairs based on the global image posture and the image acquisition parameters to obtain a three-dimensional point cloud of the measured building area.

[0089] In a specific application scenario, in order to determine multiple three-dimensional Gaussian ellipsoids corresponding to the measured building area, the processing unit 33 also includes a decomposition module 334.

[0090] The second determination module 332 may also be used to determine the three-dimensional coordinates of each three-dimensional point in the three-dimensional point cloud, and determine the three-dimensional point color and the three-dimensional point opacity of each three-dimensional point.

[0091] The second determination module 332 may also be used to determine the point cloud center of the three-dimensional point cloud, and determine the covariance matrix corresponding to each of the three-dimensional points based on the distance and direction from each of the three-dimensional points to the point cloud center.

[0092] The decomposition module 334 may be used to perform eigenvalue decomposition on the covariance matrix to obtain a rotation matrix and a scaling matrix corresponding to each of the three-dimensional points.

[0093] The second determination module 332 can also be used to determine the three-dimensional Gaussian ellipsoid corresponding to each of the three-dimensional points based on the three-dimensional coordinates corresponding to each of the three-dimensional points, the three-dimensional point color, the three-dimensional point opacity, the covariance matrix, the rotation matrix, and the scaling matrix, wherein the ellipsoid center of the three-dimensional Gaussian ellipsoid is the three-dimensional coordinate of the corresponding three-dimensional point, the covariance matrix of the three-dimensional point determines the shape of the corresponding three-dimensional Gaussian ellipsoid, the rotation matrix of the three-dimensional point determines the direction of the corresponding three-dimensional Gaussian ellipsoid, the scaling matrix of the three-dimensional point determines the size of the corresponding three-dimensional Gaussian ellipsoid, the three-dimensional point color of the three-dimensional point determines the ellipsoid color of the corresponding three-dimensional Gaussian ellipsoid, and the three-dimensional point opacity of the three-dimensional point determines the opacity of the corresponding three-dimensional Gaussian ellipsoid.

[0094] In a specific application scenario, in order to construct a dense three-dimensional point cloud, the device further includes: a judgment unit 35 , a determination unit 36 ​​, and a fitting unit 37 .

[0095] The judging unit 35 may be configured to judge whether the point cloud density of the three-dimensional point cloud is less than a preset density threshold. If it is greater than or equal to the preset density threshold, the three-dimensional point cloud is determined to be a dense three-dimensional point cloud.

[0096] The determination unit 36 ​​may be configured to determine each reference point within a preset local neighborhood corresponding to each 3D point in the 3D point cloud if the density is less than the preset density threshold.

[0097] The fitting unit 37 can be used to fit a local surface in a corresponding preset local neighborhood based on each of the reference points using a preset weighted least squares method, wherein the geometric similarity between the local surface and the actual surface formed by each of the reference points in the corresponding preset local neighborhood meets preset requirements.

[0098] The determining unit 36 ​​may be specifically configured to determine the number of three-dimensional points in the three-dimensional point cloud, and based on the number of three-dimensional points, randomly generate multiple groups of external points in an external three-dimensional space of the three-dimensional point cloud.

[0099] The determination unit 36 ​​can be specifically used to project each group of external points onto the local surface corresponding to the corresponding three-dimensional point, obtain the local surface projection point corresponding to each of the three-dimensional points, and determine each of the local surface projection points as a newly added three-dimensional point. The three-dimensional point cloud and the newly added three-dimensional points constitute a dense three-dimensional point cloud.

[0100] The processing unit 33 may be specifically configured to perform Gaussian distribution processing on the dense three-dimensional point cloud to obtain a plurality of three-dimensional Gaussian ellipsoids corresponding to the measured building area.

[0101] In a specific application scenario, in order to optimize the three-dimensional Gaussian ellipsoid, the device further includes: an optimization unit 38.

[0102] The optimization unit 38 can be used to construct a parameter optimization bounding box of a three-dimensional Gaussian ellipsoid based on the position of each three-dimensional point in the three-dimensional point cloud in the three-dimensional coordinate system; define each of the images as a reference image, and based on the shooting angle corresponding to each of the reference images, project each of the three-dimensional Gaussian ellipsoids onto a projection plane corresponding to each of the shooting angles to obtain a projection image corresponding to each of the reference images; compare the projection image with the corresponding reference image for similarity, and determine the parameter optimization function corresponding to each of the three-dimensional Gaussian ellipsoids based on the comparison result; optimize the parameters of the three-dimensional Gaussian ellipsoid within the parameter optimization bounding box based on the parameter optimization function to obtain a three-dimensional Gaussian ellipsoid after optimizing the parameters; wherein the parameters include the direction, size, color, and opacity of the three-dimensional Gaussian ellipsoid. ; Input each of the images into the large model for depth estimation to obtain a predicted depth information map of the measured building area; determine the top-view projection plane of the measured building area, project each of the three-dimensional Gaussian ellipsoids after the optimization parameters onto the top-view projection plane to obtain a top-view projection image; determine the depth value of each pixel in the top-view projection image, and determine the top-view depth information map of the measured building area based on the depth value of each pixel; determine the position loss function of the three-dimensional Gaussian ellipsoid after each of the optimization parameters based on the predicted depth information map and the top-view depth information map, and based on the position loss function, optimize the ellipsoid center of each of the three-dimensional Gaussian ellipsoids after the optimization parameters in the parameter optimization bounding box, and replace the three-dimensional Gaussian ellipsoid with the three-dimensional Gaussian ellipsoid after position optimization.

[0103] In a specific application scenario, in order to determine the orthophoto of the measured building area, the image generation unit 34 includes a sorting module 341 and a projection module 342 .

[0104] The sorting module 341 may be configured to sort the three-dimensional Gaussian ellipsoids based on the depth information to obtain sorted three-dimensional Gaussian ellipsoids, and determine pixel projection weights of the three-dimensional Gaussian ellipsoids based on the depth information.

[0105] The projection module 342 can be used to use a rasterization algorithm to project the sorted surface points on each of the three-dimensional Gaussian ellipsoids to the two-dimensional projection plane according to position and pixel, and based on the pixel projection weight, superimpose the surface points on each of the three-dimensional Gaussian ellipsoids onto the pixel values ​​on the two-dimensional projection plane to obtain an orthophoto of the measured building area.

[0106] It should be noted that for other corresponding descriptions of the functional modules involved in the orthophoto generation device provided in the embodiment of the present invention, reference can be made to Figure 1 The corresponding description of the method shown will not be repeated here.

[0107] Based on the above Figure 1 The method shown, accordingly, an embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the following steps are implemented: in response to an orthophoto image generation signal of the measured building area, a multi-view image obtained by shooting the measured building area from multiple perspectives is obtained, and the image acquisition parameters of the image acquisition device corresponding to the multi-view image are determined; feature points with significant geometric structures in each of the images are extracted, cross-image matching is performed on the feature points in each of the images, and based on the matching results, feature point pairs with similar geometric structures are determined in each of the images; based on the feature point pairs and the image acquisition parameters, three-dimensional reconstruction is performed on the regional points in the measured building area represented by the feature point pairs to obtain a three-dimensional point cloud of the measured building area, and Gaussian distribution processing is performed on the three-dimensional point cloud to obtain multiple three-dimensional Gaussian ellipsoids corresponding to the measured building area; a two-dimensional projection plane of the measured building area is determined, and depth information of each of the three-dimensional Gaussian ellipsoids from the two-dimensional projection plane is determined, and based on the depth information, each of the three-dimensional Gaussian ellipsoids is splashed into the two-dimensional projection plane to obtain an orthophoto of the measured building area.

[0108] Based on the above Figure 1 The method shown and Figure 3 The embodiment of the device shown in the figure, the embodiment of the present invention also provides a physical structure diagram of a computer device, such as Figure 5As shown, the computer device includes: a processor 41, a memory 42, and a computer program stored in the memory 42 and executable on the processor, wherein the memory 42 and the processor 41 are both arranged on a bus 43, and when the processor 41 executes the program, the following steps are implemented: in response to an orthophoto image generation signal of a detected building area, a multi-view image obtained by photographing the detected building area from multiple viewing angles is acquired, and image acquisition parameters of an image acquisition device corresponding to the multi-view image are determined; feature points with significant geometric structures in each of the images are extracted, cross-image matching is performed on the feature points in each of the images, and based on the matching results, the feature points in each of the images are matched. Determine feature point pairs with similar geometric structures in the image; based on the feature point pairs and the image acquisition parameters, perform three-dimensional reconstruction on the regional points in the measured building area represented by the feature point pairs to obtain a three-dimensional point cloud of the measured building area, and perform Gaussian distribution processing on the three-dimensional point cloud to obtain a plurality of three-dimensional Gaussian ellipsoids corresponding to the measured building area; determine the two-dimensional projection plane of the measured building area, and determine the depth information of each of the three-dimensional Gaussian ellipsoids from the two-dimensional projection plane, and based on the depth information, splash each of the three-dimensional Gaussian ellipsoids into the two-dimensional projection plane to obtain an orthophoto of the measured building area.

[0109] Through the technical solution of the present invention, the present invention obtains multi-view images by photographing the measured building from multiple perspectives, and determines the feature point pairs with similar geometric structures in each image, and then performs three-dimensional reconstruction on the regional points in the measured building area based on the feature point pairs and image acquisition parameters to obtain a three-dimensional point cloud. In the process of orthophoto generation, by constructing a three-dimensional point cloud, more accurate and detailed building information, including occlusion, can be obtained, so that when generating an orthophoto, the real form and structural characteristics of the building can be more accurately reflected, thereby improving the generation accuracy of the orthophoto, and then the three-dimensional point cloud is processed The method comprises the following steps: first, forming a three-dimensional Gaussian ellipsoid, and then splashing each three-dimensional Gaussian ellipsoid onto a two-dimensional projection plane corresponding to the measured building to obtain an orthophoto. In the process of generating an orthophoto, the present invention constructs a three-dimensional Gaussian ellipsoid of the measured building. Since the three-dimensional Gaussian ellipsoid can accurately represent the three-dimensional form of the building and the three-dimensional Gaussian ellipsoid is editable, the shape of the building can be fine-tuned as needed to adapt to different application scenarios. At the same time, the three-dimensional Gaussian ellipsoid can maintain consistency under different viewing angles, thereby enhancing the stereoscopic sense and authenticity of the orthophoto, so as to improve the generation quality and generation accuracy of the orthophoto.

[0110] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a 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 can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0111] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for generating orthophotos based on three-dimensional Gaussian splashing, characterized in that: include: In response to an orthophoto generation signal of the measured building area, a multi-view image is obtained by photographing the measured building area from multiple viewing angles, and an image acquisition parameter of an image acquisition device corresponding to the multi-view image is determined; Extracting feature points with significant geometric structures in each of the images, performing cross-image matching on the feature points in each of the images, and determining feature point pairs with similar geometric structures in each of the images based on the matching results; Based on the feature point pairs and the image acquisition parameters, three-dimensional reconstruction is performed on the regional points in the measured building area represented by the feature point pairs to obtain a three-dimensional point cloud of the measured building area, and it is determined whether the point cloud density of the three-dimensional point cloud is less than a preset density threshold. If it is greater than or equal to the preset density threshold, the three-dimensional point cloud is determined to be a dense three-dimensional point cloud; if it is less than the preset density threshold, each reference point in a preset local neighborhood corresponding to each three-dimensional point in the three-dimensional point cloud is determined; based on each of the reference points, a local surface is fitted in the corresponding preset local neighborhood using a preset weighted least squares method, wherein the local surface is consistent with the corresponding The geometric similarity of the actual surface formed by each of the reference points in a preset local neighborhood meets the preset requirements; the number of three-dimensional points in the three-dimensional point cloud is determined, and based on the number of three-dimensional points, a plurality of groups of external points are randomly generated in the external three-dimensional space of the three-dimensional point cloud; each group of external points is projected onto the local surface corresponding to the corresponding three-dimensional point, and the local surface projection point corresponding to each of the three-dimensional points is obtained, and each of the local surface projection points is determined as a newly added three-dimensional point, and a dense three-dimensional point cloud is formed by the three-dimensional point cloud and the newly added three-dimensional points; the dense three-dimensional point cloud is subjected to Gaussian distribution processing to obtain a plurality of three-dimensional Gaussian ellipsoids corresponding to the measured building area; Determine the two-dimensional projection plane of the measured building area, and determine the depth information of each three-dimensional Gaussian ellipsoid from the two-dimensional projection plane. Based on the depth information, splash each three-dimensional Gaussian ellipsoid into the two-dimensional projection plane to obtain an orthophoto of the measured building area.

2. The orthophoto generation method based on three-dimensional Gaussian splatting according to claim 1, characterized in that: The step of extracting feature points having significant geometric structures from each of the images comprises: Taking each pixel point in each of the images as a pixel point to be processed, and determining the pixel mean value of each pixel point in a preset neighborhood corresponding to each pixel point to be processed in each of the images; Replacing the pixel values ​​of corresponding pixels to be processed with the pixel mean values ​​to obtain the preprocessed images; The preprocessed images are respectively used as the bottommost pyramid layer images of the corresponding image pyramid to be constructed; Each pyramid layer except the bottom pyramid layer in the image pyramid to be constructed is respectively used as a target pyramid layer, and based on the number of rows and columns of the image in the upper pyramid layer corresponding to the target pyramid layer, Gaussian blur processing and downsampling processing are sequentially performed on the image in the upper pyramid layer to obtain the target pyramid layer image, and based on the bottom pyramid layer image and each of the target pyramid layer images, an image pyramid corresponding to each of the images is constructed; The images in two adjacent pyramid layers in the image pyramid are processed into images with the same resolution, and the images with the same resolution in the two adjacent pyramid layers are subtracted to obtain a Gaussian difference pyramid, wherein the method for processing the images in two adjacent pyramid layers in the image pyramid into images with the same resolution comprises: determining a low-resolution image layer and a high-resolution image layer in the two adjacent pyramid layers, and determining a magnification factor of the image in the low-resolution image layer based on the number of rows and columns of the image in the high-resolution image layer and the number of rows and columns of the image in the low-resolution image layer; Based on the magnification factor, up-sampling the image in the low-resolution image layer to obtain an image with the same resolution that matches the resolution of the image in the high-resolution image layer; The difference pyramid layers except the first layer and the last layer in the Gaussian difference pyramid are respectively used as a target difference pyramid layer, and each pixel point of the image in the target difference pyramid layer is respectively used as a target pixel point; Determine, in a difference pyramid layer adjacent to the target difference pyramid layer, each adjacent evaluation pixel point in a preset difference neighborhood corresponding to the target pixel point, and determine, in the target difference pyramid layer, each same-layer evaluation pixel point in a preset same-layer neighborhood corresponding to the target pixel point; If the target pixel point is the maximum pixel value point or the minimum pixel value point among the adjacent evaluation pixel points and the same-layer evaluation pixel points, the target pixel point is determined as a feature point with a significant geometric structure in the corresponding image.

3. The orthophoto generation method based on three-dimensional Gaussian splatting according to claim 1, characterized in that: The method of performing three-dimensional reconstruction on the regional points in the measured building area represented by the feature point pairs based on the feature point pairs and the image acquisition parameters to obtain a three-dimensional point cloud of the measured building area includes: Based on the feature point pairs, an image association graph corresponding to each of the images is generated, wherein a node in the image association graph represents each of the images, and an edge in the image association graph represents a geometric relationship between two connected images, wherein the geometric relationship refers to a relative position and a difference in viewing angle between the two images obtained through the feature point pairs; Based on the geometric relationship between the two connected images in the image association graph, determining the basic matrix and the essential matrix between the two connected images; Based on the basic matrix and the essential matrix, determine the relative posture between the two connected images, and based on the relative posture between each of the two connected images, determine the global image posture corresponding to each of the images, wherein the relative posture is the relative rotation matrix and the translation vector between the two connected images; Based on the global image posture and the image acquisition parameters, three-dimensional reconstruction is performed on the regional points in the measured building area represented by the feature point pairs to obtain a three-dimensional point cloud of the measured building area.

4. The method for generating orthophotos based on three-dimensional Gaussian splatting according to claim 1, characterized in that: The performing Gaussian distribution processing on the three-dimensional point cloud to obtain a plurality of three-dimensional Gaussian ellipsoids corresponding to the measured building area includes: Determining the three-dimensional coordinates of each three-dimensional point in the three-dimensional point cloud, and determining the three-dimensional point color and the three-dimensional point opacity of each three-dimensional point; Determine the point cloud center of the three-dimensional point cloud, and determine the covariance matrix corresponding to each of the three-dimensional points based on the distance and direction from each of the three-dimensional points to the point cloud center; Performing eigenvalue decomposition on the covariance matrix to obtain a rotation matrix and a scaling matrix corresponding to each of the three-dimensional points; Based on the three-dimensional coordinates, the three-dimensional point colors, the three-dimensional point opacity, the covariance matrix, the rotation matrix, and the scaling matrix corresponding to each of the three-dimensional points, the three-dimensional Gaussian ellipsoid corresponding to each of the three-dimensional points is determined, wherein the ellipsoid center of the three-dimensional Gaussian ellipsoid is the three-dimensional coordinates of the corresponding three-dimensional point, the covariance matrix of the three-dimensional point determines the shape of the corresponding three-dimensional Gaussian ellipsoid, the rotation matrix of the three-dimensional point determines the direction of the corresponding three-dimensional Gaussian ellipsoid, the scaling matrix of the three-dimensional point determines the size of the corresponding three-dimensional Gaussian ellipsoid, the three-dimensional point color of the three-dimensional point determines the ellipsoid color of the corresponding three-dimensional Gaussian ellipsoid, and the three-dimensional point opacity of the three-dimensional point determines the opacity of the corresponding three-dimensional Gaussian ellipsoid.

5. The method for generating orthophotos based on three-dimensional Gaussian splatting according to claim 1, characterized in that: After performing Gaussian distribution processing on the three-dimensional point cloud to obtain a plurality of three-dimensional Gaussian ellipsoids corresponding to the measured building area, the method further includes: Based on the position of each three-dimensional point in the three-dimensional point cloud in the three-dimensional coordinate system, construct a parameter-optimized bounding box of the three-dimensional Gaussian ellipsoid; Each of the images is defined as a reference image, and based on the shooting angle corresponding to each of the reference images, each of the three-dimensional Gaussian ellipsoids is projected onto a projection plane corresponding to each of the shooting angles to obtain a projection image corresponding to each of the reference images; Comparing the projection image with the corresponding reference image for similarity, and determining the parameter optimization function corresponding to each of the three-dimensional Gaussian ellipsoids based on the comparison result; Based on the parameter optimization function, the parameters of the three-dimensional Gaussian ellipsoid are optimized within the parameter optimization bounding box to obtain a three-dimensional Gaussian ellipsoid with optimized parameters; wherein the parameters include the direction, size, color, and opacity of the three-dimensional Gaussian ellipsoid; Input each of the images into a large model to perform depth estimation, and obtain a predicted depth information map of the measured building area; Determine a top-view projection plane of the measured building area, and project the three-dimensional Gaussian ellipsoid after each of the optimized parameters onto the top-view projection plane to obtain a top-view projection image; Determine the depth value of each pixel in the overhead projection image, and determine the overhead depth information map of the measured building area based on the depth value of each pixel; Based on the predicted depth information map and the top-view depth information map, the position loss function of the three-dimensional Gaussian ellipsoid after each optimization parameter is determined, and based on the position loss function, the ellipsoid center of the three-dimensional Gaussian ellipsoid after each optimization parameter is optimized within the parameter optimization bounding box, and the three-dimensional Gaussian ellipsoid is replaced with the three-dimensional Gaussian ellipsoid after position optimization.

6. The method for generating orthophotos based on three-dimensional Gaussian splatting according to claim 1, characterized in that: The method of splashing each of the three-dimensional Gaussian ellipsoids onto the two-dimensional projection plane based on the depth information to obtain an orthophoto of the measured building area includes: Based on the depth information, sorting the three-dimensional Gaussian ellipsoids to obtain sorted three-dimensional Gaussian ellipsoids, and based on the depth information, determining pixel projection weights of the three-dimensional Gaussian ellipsoids; A rasterization algorithm is used to project the sorted surface points on each of the three-dimensional Gaussian ellipsoids onto the two-dimensional projection plane according to position and pixel, and based on the pixel projection weight, the surface points on each of the three-dimensional Gaussian ellipsoids are projected onto the pixel values ​​on the two-dimensional projection plane for superposition to obtain an orthophoto of the measured building area.

7. An orthophoto generation device based on three-dimensional Gaussian splashing, characterized in that: include: an acquisition unit, configured to generate a signal in response to an orthophoto of a detected building area, acquire a multi-view image obtained by photographing the detected building area from multiple viewing angles, and determine an image acquisition parameter of an image acquisition device corresponding to the multi-view image; A matching unit, configured to extract feature points with significant geometric structures in each of the images, perform cross-image matching on the feature points in each of the images, and determine feature point pairs with similar geometric structures in each of the images based on the matching results; A processing unit is used to perform three-dimensional reconstruction on the regional points in the measured building area represented by the feature point pair based on the feature point pair and the image acquisition parameters to obtain a three-dimensional point cloud of the measured building area, and to determine whether the point cloud density of the three-dimensional point cloud is less than a preset density threshold; if it is greater than or equal to the preset density threshold, the three-dimensional point cloud is determined to be a dense three-dimensional point cloud; if it is less than the preset density threshold, each reference point in a preset local neighborhood corresponding to each three-dimensional point in the three-dimensional point cloud is determined; based on each of the reference points, a local surface is fitted in the corresponding preset local neighborhood using a preset weighted least squares method, wherein the local surface The geometric similarity between the face and the actual surface formed by each of the reference points in the corresponding preset local neighborhood meets the preset requirements; the number of three-dimensional points in the three-dimensional point cloud is determined, and based on the number of three-dimensional points, multiple groups of external points are randomly generated in the external three-dimensional space of the three-dimensional point cloud; each group of external points is projected onto the local surface corresponding to the corresponding three-dimensional point, and the local surface projection point corresponding to each of the three-dimensional points is obtained, and each of the local surface projection points is determined as a newly added three-dimensional point, and a dense three-dimensional point cloud is formed by the three-dimensional point cloud and the newly added three-dimensional points; the dense three-dimensional point cloud is subjected to Gaussian distribution processing to obtain multiple three-dimensional Gaussian ellipsoids corresponding to the measured building area; The image generation unit is used to determine the two-dimensional projection plane of the measured building area and determine the depth information of each three-dimensional Gaussian ellipsoid from the two-dimensional projection plane. Based on the depth information, each three-dimensional Gaussian ellipsoid is splashed into the two-dimensional projection plane to obtain an orthophoto of the measured building area.

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

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, the steps of the method according to any one of claims 1 to 6 are implemented.

Images