Research Of Point Cloud Algorithm Based On Feature Points Extraction

Three-dimensional laser scanning technology is a new data acquisition technology,which plays an important role in heritage restoration, digital city, reverse engineering.However, we can only get a perspective of the data object in the actual measurement by using three-dimensional laser scanner. So, the point cloud data registration is extremely important as 3D point cloud data processing part,and it can get a different perspective of the data through a series of calculations to convert to the same coordinate system in order to gain a complete point cloud data objects.Whether the point cloud data registration is good or bad will directly affect other subsequent processing.The traditional point cloud registration method has existed the question that was excessive consumption of time, and the registration efficiency was not high,so that,the focus of this paper is study to compare and analyze the results and the effect of different methods to extract feature points and the results of registration after extracting feature points.The feature points of the point cloud data is a basic primitives that retained the geometric properties and texture characteristics,and change the coordinate system does not make it change.And though different ways to extract the feature points of the point cloud data,could retain the geometric feature point clouds in the premise that reduced the amount of calculation. The main research results of this article are as follows:(1) Describes the commonly used methods of the initial point cloud registration and the traditional accurate registration, analyze and compare that.(2) Extracts the feature points of the discus and the model of the dargon by calculated the normal vectors、curvature and volume integral invariant, further analysis of its results,study the advantages、disadvantages and applicability of the three methods.(3) To discuss the methods of point cloud feature point,extracts the feature points of the initial point cloud after registration by the normal vectors、curvature and volume integralinvariant,then uses ICP algorithm to make the point cloud that extracted the feature points accurate registration.Comparise and analyze the result with the traditional ICP algorithm’s result. Explore deeper for the applicability of this way according to the different number of the point cloud at the same time.