| Reverse engineering, as one of an important way for the digestion, absorption and rapid development of products, has been drawn more and more attention recently. Scattered point-cloud model which obtained from 3D Scanner can be used to reconstruct CAD Model through reverse engineering. Regarding to scattered cloud data of complex model, there is still inflexible, low degree of automation, slow computation speed and insufficient accuracy issues for data segmentation and feature extraction technology. This thesis does theoretical research and experimental verification from the aspects of K neighborhood searching, estimation of differential information, data segmentation and feature extraction by combining theory and practice methods.The main research contents are as follows:Firstly, an improved point-cloud K neighborhood searching algorithm is put forward in this thesis to deal with the problem of easy leaving out local feature points when a regular grid search spot searches the point of K neighborhood. To realize the self-adapted searching in sample spots, the octree thought is combined in threedimensional grid space division firstly. Then the second division will be done basing on the number of small point-clouds originally divided in three-dimensional grid and the ‘point-cloud threshold’ to determine the length of raster Ln in self-adaptive space. Also, an approximate density of sampling points are self-adapted to determine the initial dynamic radius r and dynamic ball circumscribed cube to improve the space ball algorithm.Secondly, in most cases, the errors existed in normal vector calculation have impact on the estimation of curvature, which also make it difficult to estimate the curvature in regions with high curvature. The moving least-squares estimation of curvature based on rubust statistics is put forward in this thesis to estimate the curvature. Through estimation of the maximum kernel density of variable bandwidth, the optional child set of points can be got, which can be used to fit out the optimal moving leastsquares surface to calculate the curvature of surface.Thirdly, as the precision and effect of characteristics line of b-spline curve fitting is largely deviated from the original physical one, this thesis discusses a control point method to construct 3 B spline fitting characteristic line. In order to improve the precision of fitting characteristic lines, theoretical control points are directly interpolated with real control points. And accumulated chord length parametric method is adopted to Calculate the node vector. Then Choosing different K values and angle threshold values gets the optimal extraction effect.Fourthly, this thesis presents the method of mixed segmentation with the combination of kinds of cluster to handle the problem of data partitioning error which occurred in characteristics line(high curvature). With the improved k-means clustering algorithm and the mean shift algorithm based on gaussian mapping respectively on the flat area and characteristics of point cloud region segmentation, join the genetic algorithm of k-means clustering algorithm to avoid divided into local minima; In view of the characteristics of regional, using to calculate the unit normal vector of the gauss map, on the unit sphere gaussian is formed, combining adaptive mean shift method of gaussian clustering segmentation, according to the gauss map, the corresponding relationship between the point cloud data by gaussian figure after the split point cloud segmentation implementation part.Fifth, combined with Microsoft visual studio2010 and Open GL programming on the above theory, then use point cloud model of mechanical parts and hybrid as experimental object to study, the corresponding experimental results are obtained. Analysis of experimental results, and comparing with other classic algorithms, verify the k neighborhood search, differential information estimation, effectiveness and applicability of the feature extraction and data segmentation algorithm. |
PDF Full Text Request