CN-122023641-A - Three-dimensional reconstruction method for fine satellites of large-scale city building based on priori guided iteration

CN122023641ACN 122023641 ACN122023641 ACN 122023641ACN-122023641-A

Abstract

The invention discloses a three-dimensional reconstruction method of a large-scale city building fine satellite based on priori guided iteration, which comprises the steps of obtaining three data of a region to be built, namely an image sequence of full-color satellite optical image, DEM (digital elevation model) and building vector data with building layer height information, selecting one image from the image sequence as a reference image, obtaining candidate image sequences by adopting an angle screening method for subsequent calculation, resampling gray level information of the reference image to a space of a object to obtain a gray level base map, registering the building vector data with the layer height information with the gray level base map, then fusing the building vector data with the DEM resampled to the space of the object to generate elevation priori, constructing an image pyramid by gradually downsampling the candidate image sequences, and performing layer-by-layer iterative matching on the image pyramid by using an improved semi-global dense matching method of the object to obtain a reconstructed DSM.

Inventors

PAN HAIYAN
WANG JING
YANG SHUHU
XU LIJUN
HONG ZHONGHUA
Zheng Panyi
ZHOU RUYAN
TAO JIANG
JIANG CHENCHEN
PENG BO
ZHANG YUN
HAN YANLING

Assignees

上海海洋大学

Dates

Publication Date: 20260512
Application Date: 20251215

Claims (8)

1. A three-dimensional reconstruction method of a fine satellite of a large-scale city building based on priori guided iteration is characterized by comprising the following steps: step one, acquiring three data of a region to be built, namely an image sequence of full-color satellite optical images, DEM (digital elevation model) and building vector data with building layer height information, selecting one image as a reference image, and acquiring candidate image sequences by adopting an angle screening method for subsequent calculation, wherein the other images are all source images; resampling gray information of a reference image to an object space to obtain a gray base map, registering building vector data with layer height information with the gray base map, and then fusing the building vector data with the gray base map with the DEM resampled to the object space to generate an elevation priori; Step three, gradually downsampling a candidate image sequence to construct an image pyramid, and then performing layer-by-layer iterative matching on the image pyramid from top to bottom by using an improved object space semi-global dense matching method to obtain a reconstructed DSM; The improved object space semi-global dense matching method adopts CENSUS to calculate cost value, then adopts a cross arm cost aggregation method guided by a building prior and a scanning line cost aggregation method guided by a gray base map to conduct two-stage aggregation optimization on the cost value, and when iterative matching is conducted on an image pyramid layer by layer, a Gao Chengsou element range of the topmost layer is initialized by using a high Cheng Xianyan, and matching results of layers above Gao Chengsou element ranges of the other layers constraint a high Cheng Sousuo range of the lower layer.
2. The three-dimensional reconstruction method of the fine satellite of the large-scale city building based on priori guided iteration of claim 1, wherein when the image pyramid is constructed, the source image to be matched, the reference image and the corresponding object space are subjected to step-by-step downsampling, and the sampling interval of each layer except the full-size layer at the bottommost part is executed according to the following relation; And (3) image: , ; Object space: , , ; Wherein, the And V represents the displacement steps in the horizontal and vertical directions of the image, , Y, Z respectively represents the displacement step length of the longitude and latitude direction and the height direction in the object space, A layer sequence number is represented; For the full-size layer, namely the 0 th layer, the image is not subjected to downsampling, and the displacement step relation of the object space is set as , , 。
3. The three-dimensional reconstruction method of the fine satellite of the large city building based on priori guided iteration of claim 2, wherein an improved object-space semi-global dense matching method is adopted to match the current layer of the image pyramid when layer-by-layer iteration Matching to obtain elevation values corresponding to the horizontal coordinates v of all object points Together forming an elevation map Then to the elevation map Performing morphological etching and morphological expansion to obtain elevation values And Then the next layer of the image pyramid The search range corresponding to the horizontal coordinate v is expressed as ) , wherein, For the step length of the elevation displacement, As a function of the step size method coefficients, For full-size layers, its search range represents ) , wherein, Is the elevation value after matching with the upper layer adjacent to the full-size layer.
4. The three-dimensional reconstruction method of fine satellites for large urban architecture based on priori guided iteration of claim 3 wherein the search range of the topmost layer of the image pyramid is defined according to an elevation priori, the elevation priori being divided into ground elevations Building contour buffer elevation And building elevation within the building contour Three types of the liquid are used, Ground elevation Is the lowest elevation of (2) And the highest elevation The definition is as follows: Wherein, the Elevation values from a reference DEM image; Building contour buffer elevation Is the lowest elevation of (2) And the highest elevation The definition is as follows: Wherein, the From a reference building height; building elevation within building contour buffer Is the lowest elevation of (2) And the highest elevation The definition is as follows:
5. The three-dimensional reconstruction method of large-scale city building fine satellite based on priori guided iteration of claim 1, wherein for the improved object space semi-global dense matching method, the cost value is calculated by adopting the following formula , Wherein, the Representing the transformation value corresponding to the gray value in the projection window on the reference image, Representing the transformation value corresponding to the gray value in the projection window on the ith Zhang Yuan images, N is the total number of source images, Representing the calculated cost value of the i Zhang Yuan th image and the reference image, The weighted average value representing all the cost values is the final cost value, wherein u and v represent the coordinates of the vertical and horizontal directions in the image, and X, Y and Z represent the coordinates of any object point in the object space in the longitudinal direction, the latitudinal direction and the elevation direction respectively; in using cross arm cross cost aggregation, an object point V in a given object space When it finds an endpoint When it stops the arm in a certain direction, the stopping rule at this point is as follows: Wherein, the Representing the color difference of two object points, the corresponding gray information comes from the gray base map, And The gray-level threshold is represented and, Representing a more stringent gray scale constraint and, Representing the spatial distance of two object points, And Representing the distance threshold value(s), Representing the previous object point, in the case of the left arm, Indicating the extent of the roof of the building, Represents a ground range; the calculation formula of the scan line cost aggregation is as follows: Wherein, the Representing the coordinates of an object point V in object space, Equivalent to the horizontal coordinates (X, Y) of the object point, Z represents the coordinates of the object point in the elevation direction, Representing the cost value of the object point V after the scan line cost aggregation assuming the elevation Z, Representing object square points after cross arm cross cost aggregation The cost value, r defines the aggregation direction, And Represents penalty term, satisfy , Representation of Corresponding to the small parallax change and the large parallax change respectively, And Respectively represent object points in object space And Is in gray-scale base map Corresponding to the gray value of the color filter.
6. The three-dimensional reconstruction method of fine satellites for large urban architecture based on priori guided iteration of claim 1, wherein when a candidate image sequence is obtained by adopting an angle screening method in the first step, selecting an image with the smallest solar zenith angle and the imaging angle closest to a vertical viewing angle from the image sequence as a reference image, taking the rest images as source images, then calculating intersection angles and solar angle differences between the reference image and each source image one by one, and finally selecting the source images with intersection angles and solar angle differences larger than a set threshold value and the reference images to form the candidate image sequence.
7. The three-dimensional reconstruction method of fine satellites for large urban architecture based on priori guided iteration of claim 6 wherein the intersection angle between the reference image and any source image is calculated by using the following formula And solar angle difference , Wherein, the And The principal point vector angles of the reference image and the source image are respectively represented, And Respectively representing the solar zenith angle and solar azimuth angle of the reference image, And The solar zenith angle and solar azimuth angle of the source image are represented, respectively.
8. The three-dimensional reconstruction method of the large-scale city building fine satellite based on priori guided iteration of claim 1, wherein in the second step, building vector data with layer height information is firstly divided into a plurality of small blocks uniformly, then local affine or homography transformation is used for registering the corresponding area of each small block and a gray base map, and then registration results are rasterized and resampled to object space together with DEM for superposition fusion to generate elevation priors.

Description

Three-dimensional reconstruction method for fine satellites of large-scale city building based on priori guided iteration Technical Field The invention belongs to the technical field of three-dimensional reconstruction, and particularly relates to a three-dimensional reconstruction method of a fine satellite of a large-scale city building based on priori guided iteration. Background With the acceleration of the global urbanization process, the rapidly growing population of cities presents challenges to the design planning of cities. The three-dimensional model generated by three-dimensional reconstruction of the urban area provides accurate spatial data support for urban fine management, and has important value in modern urban management and planning, wherein the automatic three-dimensional reconstruction method based on satellite remote sensing optical images is widely applied to the generation of the urban three-dimensional model to assist urban management due to the inherent advantages of low optical image acquisition cost and wide coverage range, and is widely focused by the photogrammetry and remote sensing world. Although the development of the conventional reconstruction method in the satellite three-dimensional reconstruction field has made it possible to automatically produce geographic three-dimensional products such as Digital Surface Models (DSM) according to stereoscopic satellite optical images, in such complex scenes as cities, due to the inclusion of a large number of high-rise buildings, complex building topologies and dense building distributions, more serious shadows, shadows and complex topologies are commonly present in urban scenes as compared with natural landform scenes, which results in the conventional reconstruction method exhibiting significant degradation of reconstruction quality in the reconstruction task of urban scenes. Although the prior art improves the traditional semi-global stereo matching algorithm in the field of computer vision aiming at the characteristics of remote sensing images, a three-dimensional reconstruction frame oriented to satellite images is constructed. However, due to severe shadows and shadows caused by densely arranged building groups with complex structures and fine topologies in urban scenes, and the mixed pixel phenomenon commonly existing in the satellite image imaging process, the three-dimensional reconstruction result presents typical problems of building structure adhesion, elevation value deviation, edge blurring, reconstruction cavity and the like. As shown in fig. 1, the following problems mainly exist in a method for generating a semi-global closely matched satellite stereoscopic Digital Surface Model (DSM) based on object space as proposed by 2016 Ghuffar ‌ et al: The method of ghuffar et al adopts a constant penalty term in the cost aggregation stage of object space semi-global dense matching, namely, uniform smooth constraint is applied to any parallax change in all pixel neighbors, and the fixed penalty strategy can effectively inhibit noise in a flat or texture continuous region, but lacks self-adaptive adjustment capability to image structural features. Because the fixed punishment can not adjust the constraint intensity according to the image texture or gradient change, the method is limited in the urban area, and particularly, excessive smoothness is easy to generate at the abrupt elevation boundary of the building roof and the elevation boundary, the building and the ground, and the like, so that the building edge is fuzzy, the elevation collapse or the structure is incomplete. The method of ghuffar et al uses a smooth hypothesis based cost aggregation at the object semi-global dense matching cost aggregation stage, but the initial cost value may contain significant errors due to the common occlusion, shadowing and mixed pixel phenomena in urban areas. These errors are accumulated gradually because they cannot be effectively screened in the subsequent polymerization process, and finally, the reconstruction result cannot accurately reflect the real structural characteristics of the ground object, so that the problems of reduced quality of building edge reconstruction and misalignment of building height estimation are caused. The method of ghuffar et al uses the matching results of the low resolution pyramid to limit the parallax search range of the high resolution pyramid to speed up the efficiency of the dense matching algorithm in dense matching of large-scale images. However, in practical application of urban scenes, dense urban scenes often have the characteristics of too small space between buildings, complex building topological structure and abundant detail features, and at this time, if images are subjected to multi-layer downsampling, a great amount of detail can be lost, and meanwhile, the problems of building adhesion, building body loss, building detail loss and the like caused by uncertainty of gray information brough