Registration Technology Of Point Cloud Based On Improved ICP Algorithm

Reverse Engineering is an important designing technique, which is important toaccelerate the development of technique and science of China. Among the steps ofReverse Engineering, data-mating object, processing data, and rebuilding3-D modelare the essential steps. However, to decide whether a product satisfies the designingrequirement, the key step is Geometry quality evaluation.This topic originates from the project of the geometric quality evaluation ofturbine blade. A special coordinate measurement machine (CMM) is designed tomeasure the data of turbine blade. The point cloud of blade is collected to comparewith the standard model and evaluate the error and tolerance. The direction of thispaper is the registration and evaluation between the measuring data and the standardmodel. The following is the primary contents:This essay introduces kinds of methods of registration, and elaborates the mostpopular one, ICP (Iterative closest point) algorithm. For the defect of this algorithm,paper proposes to register point clouds roughly before precise registration. And, itintroduces the concept of nearest neighbor search to algorithm, improves searchingspeed of corresponding points using k-D tree data structure. The accuracy andefficiency of algorithm is improved through the threshold method of the directionvector of the point in the point clouds.Finally, for testing this algorithm, we design four different tests. Four differenttypes of surface, space ruled surface, scattered point cloud, dense point cloud andsparse point cloud, are used to testified this algorithm. The result shows that theaverage error is0.0320274mm, the error of the flat surface registration is0.000007mm, solving the problem of surface registration.