Point Cloud Models Segmentation Based On Bhattacharyya Distance And Shape Diameter Function

As the research foundation of 3D model retrieval,geometric deformation,3D modeling,etc.,3D point cloud segmentation has been widely used in many fields.For example,in traditional industrial manufacturing,product design are carried out by combining deformations of different segments.Development,speeding up the efficiency of product design,and in terms of medical diagnosis,by segmenting complex organs,not only narrowing the scope of diagnosis,but also separating the diseased organs from the surrounding organs to reduce the complexity of treatment.Influenced by noise points,point cloud simplification,point cloud feature estimation,etc.,existing point cloud segmentation methods are prone to boundary jagged,oversegmented or under-segmented problems.Aiming at the existing problems,this paper proposes a point cloud segmentation method based on the bhattacharyya distance and shape diameter function,and carries out corresponding experimental verification.The main work done in this paper is as follows:(1)Based on spectral clustering and K-means++ algorithm,the initial segmentation of point cloud data is realized.The under-segmentation problem in the initial segmentation results is solved by using visibility.The K-nearest neighbor algorithm is used to extract the boundary points and their neighbors,and normal vector is used to construct the histogram.Region merge is based on bhattacharyya distance.to calculate the similarity of adjacent region.(2)The order relationship analysis method is introduced to improve the calculation of the shape diameter function.The SDF value of the point is calculated by the improved shape diameter function,the SDF value is normalized and the histogram is constructed,and the similarity of the histogram is calculated by the Earth Mover's Distance.The axis and Gaussian curvature improve the Snake algorithm to smooth the boundaries.(3)The algorithm is based on MATLAB and C++,and the experimental data is validated by the data set published by Princeton University.The effectiveness and superiority of the proposed algorithm are verified by comparison with existing unsupervised algorithms.