| With the rapid development of3D scanning technologies, modeling detailed3D shapes by scanning real physical objects is becoming more and more mature, which makes point cloud become a new type of digital media data. In recent years, Point-based Graphics is playing a significant role in the international research field of computer graphic. And feature exaction and segmentation of point clouds are one of basic and key technologies in digital geometry processing, which play great practical role in surface reconstruction, shape rendering and so on. Based on the sound mathematical foundation and experiment foundation, we take a deep research on a series of issues which includes geometric analysis, point cloud simplification and feature extraction and segmentation. The major contributions of the dissertation are:The methods for computing geometric attributes are proposed. Based on principal covariance analysis over K-nearest points, normals of point cloud can be computed, then we use propagation algorithm to adjust the normal orientation. Besides, we calculate the curvatures and projection residuals of every point by applying moving least squares method.Proposed feature-preserved simplification algorithm for point cloud models. Algorithm first calculates an axis-aligned bounding box for the point sets, hierarchically and adaptively subdivides the box into eight cubes with octree data structure. Then identify the characteristic points of the point cloud by principal covariance analysis and least squares surface fitting method. On traversaling the octree structure, non-characteristic points are simplified and characteristic points are preserved.Algorithm for extracting sharp features form point clouds. Our algorithm first calculates projection residuals and identifies potential feature points in point cloud model. It then smoothes the potential feature points by employing a modified version of the principal component analysis approach. Subsequently, a feature-ployline propagation technique is used to approximate the feature points by a set of polylines. Algorithm finally optimizes the feature curves by resolving gaps and recovering junctions. Experiments show that the algorithm is very robust and it can extract feature curves from various point clouds.Based on extracted sharp features curves, an algorithm of automatic segmenting point clouds was presented in this paper. Our algorithm approximates the sharp feature points by cubic B-spline curve. Based on the constraint of optimized feature curves, region growing algorithm was applied to segment the point clouds into multiple regions, the geometric features of the region is consistent and the boundary of the patch is neat. Experiments show that the algorithm can segment the point clouds precisely and efficient. Our algorithm can be used in shape matching, texture mapping, CAD modeling and reverse engineering.The above five different research issues are independent, while are also in conjunction with each other. The experimental results demonstrated the high validity, practicality, and generality of algorithms introduced in this thesis. As deep study on related data and analysis of experiment results, we give three significant topics for future work in the conclusion chapter. |
