The Research On The Scattered Point Clouds Simplification And Automatic Registration Technology

In reverse engineering, 3D optical scanning equipment is used to measure the object’s surface, and it must be conducted from different directions for the complete data of the surface. For the position deviation between each point cloud, they need to be registered for the effective data model; Besides, there are lots of redundant data in the dense points which used to describe the surface features and reconstruct the model, therefore data simplification is necessary.The quality of scattered point cloud after simplification and registration, affects the efficiency and quality of the late model reconstruction. Therefore, this paper studied scattered point cloud simplification and automatic registration intensively; then combine VC++ 6.0 with OpenGL, developed the point cloud pretreatment software. In this paper, the research emphasis are as follows:(1) Put forward the adaptive simplification algorithm based on the minimum elliptical distance and plane-angle method used to extract characteristics of sharp boundary. According to the curvature and vector of the point cloud and based on plane-angle method and maximum angle method, to complete the segmentation of the point cloud and the extraction of the edge feature; Remain the boundary points, simplify the points of the plane region by using the minimum distance method and the points of the non-planar data points by using the adaptive simplification algorithm based on the minimum elliptical distance. To the model containing 64418 points, spends 5.65 s totally from importing point cloud to the end of the point cloud simplification, and the reduced rate is 69.3%, the grid reconstruction model compared with original model, the main distance deviation ranges from-0.11 to +0.11.(2) Improved ICP algorithm based on four points and improved ICP algorithm based on principal component analysis are presented. Firstly, extract the data points which curvatures are relatively large, to form two feature points sets; For four points algorithms, find the corresponding relationship according to the length of two diagonals and the division ratio of the node; For principal component analysis, cluster the nearby points of the seed points selected according to the invariant moment features, to form two new point sets. Then calculate the transformation parameters used to do registration by the four points-pair or the two new point sets. Finally, improve the ICP algorithm, take the closest points’ neighborhood points which characteristic are more similar instead of the closest points, and achieve precise registration. To two pieces of point clouds of toys cartridge clip(points number 93561, 97720), do registration experiments, from importing the data to finishing registration, the method based on PCA consumes 327.5s, iterate 11 times to convergence. And the method based on 4PCS consumes 291.7s, iterate 7 times to convergence.(3) Complete the point cloud pretreatment software. Using VC++ 6.0 combining with OpenGL to finish the relate function, such as the import, export, display and interaction as well as the simplification and registration of the point clouds. Finally, the examples shows that, the system can realize the various functions properly.