Research On The Algorithm For Registration Of 3D Point Cloud And CAD Model

With the development of society and technology, people need more products with higher level industrial design. And various curved surfaces and combinations of complex surfaces in mobile phones, home appliances, automobiles and other products have been widely used in the design. All these changes prompt component detection rapid transition from 2D to 3D technology. Today digital high precision detection based on CAD model has become a major trend. One of the key technology is the registration of point cloud and CAD model. More often, the coordinate system of point cloud and CAD model are different. In order to evaluate the error of workpiece, we must achieve the registration of point cloud and CAD model. Among the algorithms of the point cloud registration, the TCP and variants of ICP have been widely used. In this paper, depth analysis has found that the ICP algorithm still has limitation in the registration of point cloud and CAD model. The limitation mainly manifests in following aspects. Firstly, the original ICP algorithm is based on a basic assumption that the closest point is the corresponding point by Euclidean distance. There would be the problem that searching for corresponding points are time consuming and crude when the point cloud of CAD model are uneven or sparse. Secondly, the iterative step and process of original ICP algorithm are slow. Especially the registration efficiency is very low when there are large numbers of measuring points cloud.To solve the above problem, we proposed a modified ICP algorithm based on dynamic adjustment factor to speed up registration of point cloud and CAD model with high accuracy. The algorithm carries out the following optimizations on the base of original ICP.In the first step, we present a novel solution to search for corresponding point based on STL triangular mesh in the CAD model. The solution makes full use of the advantage that the STL file itself consists of a list of facet data and vector information of triangular mesh to calculate the corresponding point. The method avoid sampling the CAD model so as to enhance the registration efficiency. Once more, the method is a variant of ICP belongs to point-to-plane technique, so it can achieve higher accuracy than original ICP.In the second step, a modified ICP algorithm based on dynamic adjustment factor is presented in this paper based on the first step optimization. The algorithm puts forward a kind of dynamic adjustment factor-the factor of dynamically adjusting the rigid transformation parameters which can make point cloud over-travel by rigid transformation along the original trend in each iteration. When adding dynamic adjustment factor, the algorithm could search more effective corresponding points in next iteration, thus reducing the number of iterations and speeding up convergence of ICP algorithm.Experiments show that the proposed modified ICP algorithm based on dynamic adjustment factor in this paper can effectively reduce time consuming in registration of point cloud and CAD model, meanwhile improving the registration accuracy.