Study Of Segmentation Algorithm For Three-dimentional Point Cloud Data Based On Random Walks

Segmentation of point cloud plays an important role in the modification andanalysis of the existed product. The point cloud scanned by the utilization of3D digitalscanning instruments as CT scanner or MRI scanner should be preprocessed, such assegmentation of the point cloud. Then the point cloud of the part is treated withcomputer-aided design software, and the digital model of the part can be reconstructed,from which the solid part can be constructed and the mold can be designed tomanufacture the machine part. If the point cloud is fitted by a single surface or somesimple plane without the process of segmentation, the solid part may be ended with ahuge error. Therefore, this paper introduces an approach which can segment the pointcloud automatically to accomplish the reconstruction of point cloud.First, the model of random walks is deducted from the principle of potentialdifference, and the numerical solution can be calculated in this paper. By theintroduction of the voxel, the3D point cloud can be transformed to image from graph,which can be calculated in Matlab. Besides, octree is introduced to construct the k-nearest neighborhood, which can guarantee the effectiveness of the computation ofprobability in random walks. Then, the normal vector of the point cloud is approximatedby covariance analysis which can ensure the probability can be calculated effectively.At last, by the utility of the module of image processing system and mathematicalfunction library in Matlab, the result of the segmentation of point cloud can begenerated. Then, the experiments of segmentation of three types of point cloud, namelymachine part model, mixed model and model without particular character are tested.Besides, with the analysis of the threshold of probability, value of k in k-nearestneighborhood and number of the voxel, the parameters can be fixed.