Feature Extraction Of Point Cloud Based On Normal Information

Feature extraction from point cloud data is an important link of modelreconstruction in the reverse engineering of industrial parts. As rapid development ofdigital scanning technology and computer technology, three dimensional point cloudmodel has been used in reverse engineering, pattern recognition and computergraphics, etc. So the numerous scholars at home and abroad are completely interestedin feature extraction methods from point cloud. The purpose of this paper is todevelop a sort of point cloud feature extraction method.The data generated by the laser scanner or digital equipment is quite huge. Inorder to shorten the time of data processing, firstly, the point cloud data has beendivided into a lot of bounding boxes by using Octree method. Secondly, k nearestpoints can be searched in the cube of the candidate point and surrounding cubes, thenusing the least square quadric surface locally fits the k nearest points, the estimationnormalobtained bydifferential geometry knowledgeis regarded as the real normal ofthe candidate point.Finally, normal angle changes between each point and itsadjacent points identify the feature point of point cloud data.Using curve fits thesefeature points, we realize segmentation technology of point cloud data based oncharacteristics. Using identified feature points, we can divided point cloud data intofacets, then reconstruct these point cloud facets and obtain CAD model of pointcloud parts.In the Matlab environment we realize feature extraction of point cloud using theproposed feature extraction technology and verify effectiveness of this algorithm onthe actual point cloud.