Research On Ground 3D Laser Point Cloud Denoising Processing Algorithm

With the promotion and application of 3D laser scanning technology,this technology has been widely used in many fields such as industry,aerospace,medicine,film and television,archaeology,etc.Point cloud data acquisition and its processing algorithms has also attracted more and more attention from scholars at home and abroad.The point cloud data acquired by the 3D laser scanner not only stores rich information,but also has a simple structure and is easy to operate,and is suitable for expressing objects with complex surface features and irregular geometric shapes.However,affected by the accuracy of the scanning device,the experience of the operator,and external environmental factors,noise is inevitable in point cloud data,which seriously affects subsequent application processing such as point cloud registration,feature extraction,surface reconstruction,and visualization,etc.Point cloud denoising can weaken this effect and has important fundamental implications for 3D point cloud application processing.In this thesis,the point cloud obtained by the ground 3D laser scanner was taken as the research object,and the denoising process was deeply studied.The main contents are as follows:(1)The different types of laser point cloud noise and the causes of noise were summarized.At the same time,several common denoising algorithms were explained.The smoothing effect between denoising algorithms was discussed through contrast experiments.(2)Aiming at the shortcomings of existing bilateral filtering algorithlns,an improved bilateral filtering smoothing algorithm was proposed.The algorithm improves the reliability of noise position adjustment in the smoothing process based on the improvement of the quality of normal vector estimation.At the same time,sub-constraint conditions were added to the selection of point neighborhoods,which not only removes large-scale noise such as isolated points,but also enhances the robustness and security of the algorithm while improving the efficiency of algorithm.By comparing the experimental results of the bilateral filtering algorithm before and after the improvement,the superiority of the improved bilateral filtering algorithm was verified.(3)A point cloud denoising algorithm based on supervoxel segmentation was proposed.Firstly,the algorithm divides the point cloud into voxel data according to the octree.On this basis,a new similarity measure feature description operator was defined,and clustering was used to further form supervoxel data.Then,the supervoxel was modeled as a node,and the point cloud minimal spanning tree was constructed.Finally,the convex noise detection was performed according to the topological index relationship formed by the minimum tree to realize the denoising of the point cloud.