Algorithm Research Of Point Cloud Data Simplification And Registration In Three-dimensional Reconstruction Process

With the improvement of computer-aided design technology, the digital technology to collect physical data reconstruction techniques is more widely used. In this paper, the main research object is three-dimensional point cloud data. Based on extraction of the border,researching about how to simplify the point cloud data model in high precision and how to achieve accurate registration between the point cloud data, and solving the problem of improve the algorithm efficiency to some extent. In view of this, proposing a Hausdorff distance based point cloud slicing algorithm improvements to streamline and an improved ICP(Iterative Closest Point) algorithm of integrate similar principle point cloud registration. The paper’s main work is as follows:Firstly, describing the method of establishing the topology between the acquisition method of point cloud data and scattered point cloud data, including octree method, k-d tree method and rasterization method, and various methods of objective evaluation. Among them, a specific study mesh octree law principle.Secondly, proposing a point cloud simplification algorithm based on Hausdorff distance and segmentation. At first, X-Y boundary of all point cloud model is extracted to retain the appearance characteristics. Then, curvature of the remaining data points except boundary points is calculated and the fuzzy sets of point cloud model is constructed according to the average curvature. Pulling in the fuzzy set theory to get the optimal threshold of curvature segmentation. Fragmenting the point cloud data use curvature maximum and minimum difference as curvature interval one by one, so that a data point is best assigned to the current fragmentation point cloud. Finally, the Hausdorff distance of principal curvature in each fragmentation decides to get and preserve the feature points. Experiments indicate that the algorithm improve the streamlining rate, simultaneously, effectively shorten the time and canbetter reserved the detail features of point clouds.Thirdly, for the problem of registration efficiency and accuracy error existing in the process of point cloud data registration in three-dimensional reconstruction, an improved ICP algorithm is proposed. Box structure is applied to divide and sort the point cloud data, for each separate box extracting feature points to construct triangle, according to the similar principle, each vertex of the greatest similarity triangle is chosen as the initial corresponding point. In order to ensure the finding correctness of the corresponding point, introducing the concept of support degree and the evaluation criteria to guarantee the establishment of the current point to be ascertained giving the best support for the remaining matched points.Similarly, choosing every side of the existing triangle as a reference continue to build new triangle until all closest points are found. The experiment results indicate that the improved method make significantly enhance compare with traditional ICP algorithm, its algorithm advantage is obvious.