Research On Data Preprocessing And Surface Reconstruction Of Scattered Point Cloud

With the development of laser measurement technology, the surface information of an object can be accessed rapidly and accurately. However, quantity of the3D surface data are so much that it is named as "point cloud". How to process these point cloud and to generate a CAD model that can be used in manufacturing, machining or reproducing system, is one of the core task of reverse engineering. Based on that background, a series of key technologies in point cloud data processing are studied in this paper.Firstly, the normal vector of the scattered point cloud data is calculated, and the least squares algorithm is used to approximate the neighboring point set of a discrete point so that the micro-tangent plane is obtained, then, the micro-tangent plane is taken as the normal vector of the point, and its direction is adjusted to reach a global consistency.Secondly, outliers in the scattered point cloud are detected and filtered. By the analysis of the smoothing method based on the average curvature, we know that although it can remove the noise effectively, it is difficult to maintain the original characteristics of the same. So the bilateral filtering technology is applied, the characteristics of the object can be preserved due to the concept of similarity.Thirdly, scattered point cloud data block technology is studied. For determining the surface type of each point, based on the mean curvature and Gaussian curvature, the points in the scattered point cloud are classified. Next, the feature points are extracted via the adjacent corner and the curvature extreme value. Then segment the point cloud into blocks with the region growing based on the constraint of feature points.Finally, the surface reconstruction of scattered point cloud is studied. An initial triangle mesh is constructed by layer expansion. Then an accurate triangular mesh can be got by the subdivision of the initial triangle mesh. Besides, smooth the mesh vertices with bilateral filtering. The speed of this triangulation method is fast, and it can ensure that the distance from original points to the result mesh is less than the set threshold.The effectiveness of the presented approaches is demonstrated by the experimental results.