Research On Scattered Point Cloud Data Preprocessing Algorithms

Reverse Engineering(RE) is a way to convert physical models of products into digital models, and it can greatly shorten the development cycle of products.Thus, it has been widely applied in many areas. Data preprocessing is an important step of reverse engineering as it affects the quality of subsequent model reconstruction. In this thesis, some key technologies in data preprocessing are investigated.Topology relationship method for point cloud is studied. The advantages and disadvantages of several common methods are researched. This paper prop-osed an improved algorithm for building the topology relationship base on grid. The establishment of the topological relations for the point cloud data from the access to experiment. It brought convenience for point cloud preprocessing.De-noising methods for point cloud and the cause of noise are investigated.The strengths and weaknesses of several familiar de-noising methods based on the characteristics of scattered point cloud. For the lack of bilateral filtering methods, this paper proposed an improved algorithm. Then we introduce a bilateral filtering algorithm based on classification of noise in detail. The result is quite good in actual processing using this algorithm.Simplification methods for point cloud are studied. For the defectiveness of several existing methods, this paper proposed an improved algorithm based on distinguishing kind of cloud point. It provided corresponding algorithm proce-dures. Firstly, the bounding box which just contains the point cloud data was divided into several sub-spaces, and the fitting plane of K-nearest neighbor point was built by each sub-space that contains points. Secondly, the distances from each point in the K-nearest neighbor point to the fitting plane was accumulated.Thirdly, all of the accumulated distances were sorted in an ascending order, and the bounding box was divided into two domains as to ready to retain and to ready to delete. It applied the different simplification algorithms in different areas of the same point cloud data. So, the algorithm proposed in the paper can not only maintain geometric features for the point cloud data but also eliminate large holeareas more effectively. Meanwhile, the computational efficiency was increased.Registration methods for point cloud are investigated. The three-dimen-sional coordinate conversion method and the existing methods for point cloud registration are introduced. Then we propose a method for initial registration based on principal components analysis. At last, we introduce the Iterative Closest Point(ICP) algorithm.