Research On Key Issues Of UAV Dense Point Cloud Generation Based On Multiple View Geometry

The most widely used method of generating dense point cloud is based on multiple view geometry theory in the field of computer vision.The method of generating dense point cloud based on multi view geometry’s processing object is network community image with low resolution(resolution within one million),while the UAV image resolution is usually higher(in general tens of millions of levels,or even higher),so directly applied to the UAV image has some weaknesses and shortcomings.In this paper,the shortcomings of these algorithms are improved based on the characteristics of UAV images.The main work of this paper is as follows:1.In order to improve the efficiency of feature matching,retrieval method based on visual vocabulary in field of image retrieval is integrated to the UAV image matching process,and improve the original retrieval method by considering rough topology information contained in UAV images.Firstly,the original image was established in Pyramid and search for similarity ranking in the low resolution image of Pyramid.Then according to the similarity ranking establish pre-match like to narrow the search range.Finally based on the CasHash(Cascade Hash,CasHash)matching method to match in pairs.The experimental results show that the proposed algorithm significantly improves the effectiveness of UAV image matching.2.Camera global position estimation algorithm in global SfM(Structure from Motion,structure,SfM)often sensitive to outliers.First,we combined with the epipolar constraint and propose a new relative translation direction estimation method to reduce mismatch’s effect in result,reducing the outliers in the result of relative translation direction estimation.Then proposed a robust estimation model based on convex optimization subjected to L1 norm,the model solving the global camera position estimation to get a global optimal solution.The experimental results show that the proposed method is robust and accurate,and has good performance in terms of efficiency.3.In order to obtain dense point cloud model with high density,few outliers and good integrity,a new algorithm based on depth map multi view stereo reconstruction algorithm is proposed.Geometric consistency constraints are incorporated into two steps of multi view stereo reconstruction: depth map generation and depth map fusion.The specific method is to combine the optical and geometric consistency in the depth map generation process,select the matching pixel with the minimum cost which would participate in the depth value calculation,then get the depth map.In the process of fusion of depth map,firstly,the visibility and geometric and optical coherence constraints are adopted to filter outliers.The experimental results show that the proposed algorithm is robust and can get a more complete,dense model and less outlier.