Research On Compression Algorithm Of Point Cloud Data

Digital museum, as a new product produced by information society, manages and utilizes culture heritage and nature heritage from a new perspective with the renewal and development of computer science and Internet technology. Digital geological museum is able to manage large numbers of valuable geologic specimens and resources; moreover, geological resource utilization level is raised by network sharing.3D scanning and3D point cloud processing technology, which is the key technology in the realization of digital geological museum, decide whether users can get accurate models of museum exhibits via the Internet rapidly or not. The method on how to conpress point cloud data efficiently and accurately is one of the techonology approaches to get exhibits quickly to users; meanwhile, building3D models with minimum error and restrained data volumn is a significant start point. Many other fields now are facing with these two issues.In this paper, the process of3D modeling and the method to treat point cloud data aimed at the exhibits in museum was studied; research focused on point cloud data reduction algorithm and4sampling methods, radom sampling algorithm, minimum distance sampling algorithm, bounding box barycenter sampling algorithm, normal vector contained angle algorithm, was realized utilizing C#. A series of experiment on point cloud compression were done with the4methods metioned above and3D models were built with the compressed point cloud. Compressiong quality assessment of the4algorithms was carried out from three aspects, including surface area deformation, volumn deformation and standard devidation. Curvature-vetor sampling method was put forward based on the result of the quality assessment and curvature of point cloud.The main works of this paper are as follows:1. Calculation of point cloud geomatric characteristics was studied in this paper, including normal vector estimation and direction adjustment, the estimation of mean curvature and gauss curvature of point cloud based on normal vector. Calculation on spatial topological structure of point cloud was studied and point cloud was divided with spatial bound box dividing method. Based on the spatial division, K-neighborhood search of the point cloud was relized utilizing bound box extension method. All the work metioned above were the foundation of the follow-up work.2. Radom sampling algorithm, minimum distance sampling algorithm, bounding box barycenter sampling algorithm and normal vector contained angle algorithm was researched and realized, and a series experiments on data compression were done with two point cloud model, one is Venus point cloud which suface fluctuate gently, the other one is Ronaldinho toy point cloud which suface fluctuate sharply. Experiment result and visual effects of4compression method in different parameter were compared, and the relationship between compression ratio and compression parameter was studied. The curve on compression ratio VS compression parameter was obtained,3. The compression quality from the result of the4algorithms was assessed with suface area deformation, volumn deformation and standard deviation. The experimental point cloud was divided into gentle point cloud and sharp point cloud, and assessment tests were done by two representative models respectively. The experimental images were compared, and the result shows that bound box barycenter method has minimum volumn deformation while nomal vector contained angle method has the best capacity on retaining surface area characteristic and reducing the standard deviation. Comparing with volumn characteristic, standard deviation and surface area deformation is more important to measure the model error. Therefore, nomal vector contained angle method has the best compression quality than other3methods.4. Curvature-vetor sampling algorithm was proposed based on curvature calculation and nomal vector contained angle sampling. The core thought of the new method is dividing point cloud into high curvature zone and low curvature zone according to the curvature characteristic of the points, then reducing point cloud utilizing nomal vector contained angle method with different contained angle parameters. Compressiong experiment of the new reducing method with Venus point cloud and Ronaldinho point cloud was done and the compression quality was assessed. Compared with nomal vector contained angle method, the new method enhances the compression accuracy by10%-30%on gentle point cloud and by about50%on sharp point cloud. The result of actual mineral compression experiment shows that curvature-vector method is capable of dealing with real mineral point cloud model.5. Grid division, neighborhood searching, nomal vector calculation, curvature estimation and the5compression method was realized by C#coding on the.net platform.