Abstract: To improve the efficiency of reverse modeling, an automatic segmentation algorithm based on geometric attribute analysis is proposed. The algorithm subdivides point cloud into cubic grids and then maps the geometric attribute value of points in each grid to normal curvature coordinate system and Gaussian sphere. By testing hypothesis, the patterns of the normal curvature image and the Gaussian image are recognized. Based on the grid structure, the image points clustering, and the goodness-of-fit testing, point cloud is segmented into several regions and characterized as natural quadrics, extruded surfaces and ruled surfaces, respectively. Applications show that the proposed algorithm deals with large amount of measured points stably and effectively. It can be applied to many other fields including visual reality and computer vision, etc.

Key words: Point cloud, Feature extraction, Region segmentation, Reverse engineering