| Tilt photogrammetry technology,which refers to carrying single-lens or multi-lens cameras on UAV to acquire tilt and vertical images from different angles to obtain complete ground scene information with high efficiency,has become the main means of current largescale scene image acquisition and 3D reconstruction.3D reconstruction can restore UAV sequence images into 3D point clouds and realize 3D modeling of the target area,which has been extensively used in mapping,monitoring and other areas.However,during the process of tilt photography,it is necessary to acquire images of different perspectives and lighting conditions for the same target,which will encounter problems such as too large a target scale and obstacle obstruction,affecting the efficiency and accuracy of 3D reconstruction.In this paper,the key algorithms in the 3D reconstruction technology of tilt photography are studied,mainly including sparse reconstruction and dense reconstruction algorithms,and the functional implementation of the reconstructed 3D model visualization application technology.The research results and innovation points of the paper are as follows:(1)The sparse reconstruction algorithm based on incremental SFM(Structure from Motion)is studied,and the principle of the sparse matching SIFT(Scale-invariant feature transform)algorithm is introduced and improved for the problem that the algorithm takes a long time to detect feature points.By introducing the FAST(Features from accelerated segment test)algorithm to detect corner points,it can quickly determine whether a corner point is a corner point by comparing the central pixel point with the gray value of image elements in a certain domain,thus improving the speed of the algorithm.Also,to make up for the shortcomings of the FAST algorithm in terms of insufficient search on the Gaussian difference pyramid,an image segmentation algorithm based on Ostu and GA(Genetic Algorithm)is used to segment the image,construct a Gaussian pyramid for the segmented image,and perform feature point search on the Gaussian pyramid.The final processing is compared on the relevant dataset,and the validity of the algorithm is demonstrated for the bundle adjustment and sparse point cloud in the sparse reconstruction algorithm.(2)The basic principle of PMVS(Patch-based Multi-view Stereo)dense reconstruction algorithm is described,and the PMVS dense reconstruction algorithm is improved for the situation that the existing 3D reconstruction algorithm is inefficient and the reconstructed point cloud has a large number of voids.In the patch reconstruction stage,the images of polar line search and block matching are selected according to the height and overlap at the time of UAV photography,which reduces the blindness of patch reconstruction.In the patch expansion stage,the adaptive neighborhood search is used to select the diffusion range and improve the efficiency of diffusion.In the patch optimization stage,an improved proposed Newton algorithm is used to optimize the patch parameters,which improves the accuracy of relevant parameters such as patch centers and normal vectors while ensuring the global convergence of the algorithm.Finally,experiments are conducted on three data sets,showing that the improved algorithm improves the efficiency by more than 15% and reduces the number of voids in the reconstructed point cloud.(3)In the visualization application of tilt photography 3D model,a visualization platform of 3D models of tilt photography based on Cesium has been designed and implemented.On the one hand,the platform can realize the visualization display of several3 D models of houses,tall buildings and hillsides;on the other hand,it can realize layer management,spatial measurement,UAV flight path display,monomer,section analysis,slope and direction analysis and other functions.It improves the practicability of the visualization platform. |
