Study On Three-dimensional Point Cloud Registration Method Based On The Normal Vector

In recent years, reverse engineering with its inverting thinking mode has obvious advantage in shortening products development time and reducing its risks in reseaching products. It has already been one important branch of computer-aided manufacturing, and its application has also been permeated into every aspect of human life. Point cloud registration method in three-dimensional space is the key technology of reverse engineering. If we want to get the full three-dimensional object model, the multi-angle measurement is needed to be done. But we can’t remove redundant points of the data collected from different angles and assemble them into a complete three-dimensional mode without the study of point registration method.Two key problems of 3D point cloud penetrate into the entire registration algorithm’s study. The first problem is how to extract the corresponding point sets from the two collected point clouds data faster. And the second is streamlining and refining the extracted corresponding point sets and improving the matching accuracy. The main work of this thesis is as follows:First, an initial point cloud matching algorithm based on normal vector is proposed according to the geometric feature extraction method of the three-dimensional points. The normal vector of two point clouds’ points is being drawn at first, then every point’s characteristic degree can be calculated by the angle between the normal vectors. Comparing characteristic degree with threshold value set in advance, the key points can be filtered and then the curvature can be calculated. By doing this we can reduce the curvature calculation time significantly and improve efficiency of finding matching points. Key points’ initial registration has been done by the main curvature at last. The thesis verifies the performance of each algorithm from theory aspect and concrete experiments.Second, double constraints of points’ distance and the value of Gaussian curvature are established on the basis of the extracted initial matching points, the concept of equilibrium factor is also introduced. Under the influence of equilibrium factor, refining the initial matching point further and removing the similar redundant points of the main curvature’s value even the error matching points to get an exact matching point. On the basis of the exact matching point, using the method of unit quaternion to iterate in the process of minimizing objective function until get the right deviation convergence range. Then we get rotation matrix and translation matrix’s parameters between two point clouds. After doing all this, the registration is completed and the point cloud model is output at last. Finally, verifying the scientific validation and feasibility of the improved algorithm in theory aspect and experiment in chapter 5.