Research On Registration And Integration Technology For Large-Scale Point Cloud Model

With rapid development of computer technology and measurement technique, reverse engineering is popularly applied. And point cloud model registration and integration is the key technique in the reverse engineering. For the point cloud data is large-scale and noisy, general registration method can’t meet the current needs, a novel ICP algorithm based feature is present. However the point clouds can’t completely be digitalized after registration, they also need to be integrated. Therefore this paper mainly focuses on the following research.1. For the large scale and noisy point clouds,, they are denoised by two searching method of KD tree in this paper, and they are the nearest point searching and r-neighborhood searching. In addition, the point clouds are simplified by grid-based approach, they are denoised when simplified. It makes subsequent registration more successfully.2. A novel ICP algorithm based feature is present. Corresponding points are matched by relationship between surface local geometrical feature and point distance in every iteration. it makes the point cloud joining together, and the good result is achieved by this method.3. In the feature-based registration of ICP algorithm, this paper study three features in depth. The feature based on curvature statistics presents statistics of curvature which obviously changes in the local area. Three-dimensional moment group is formed to describe local characteristics of point cloud by moment principle. The feature based on normal histogram presents the relationship between normal in the local area and then to be put in the histogram. By these three features do register point clouds that they are obtained from actual measurement, finally it achieves good result. 4. The point clouds need to be integrated after registration, which makes them a totally complete point cloud model. In the paper, overlapping point is found by normal constraint; thereby the points are integrated in coordinate position and color.Compared to general registration method, the feature-based ICP registration presented in this paper has great improvement in efficiency and effect.