Research On Technology Of Point Cloud Registration And Mosaic Based On Features

With the development of computer graphics,reverse engineering has been widely used in many fields such as product manufacturing,archaeological research,virtual reality and so on.As the core technology,the point cloud registration has been deeply studied.Two main applications of point cloud registration are the quality evaluation and the model mosaic.A registration algorithm of sparse point cloud based on features is proposed with quality evaluation as the example.An algorithm of point cloud mosaic based on features is proposed with the model mosaic as example.The machining accuracy of the workpiece has a great influence on the performance of the product.As the coordinate measuring machine(CMM)has high measurement accuracy,it can be used to acquire the data of workpiece.As the point cloud acquired is sparse and has plane and cylindrical features,a registration algorithm of sparse point cloud based on features is proposed.First,the area to be registered is selected by hand and the redundant area is deleted.The plane feature is extracted using least square procedure while the cylindrical feature is extracted using nolinear-least square method of Levenberg-Marquardt.Then local coordinate system is built based on these features,rotation transformation matrix and translation vector between two point cloud is calculated and the initial registration is completed.Last,the improved ICP algorithm is applied for precise registration.Experiments are designed to verify the reliability of the algorithm.Finally,machining error of two workpieces is detected using this algorithm.The idea of the model mosaic is to find the corresponding points in the overlap region and get rigid transform matrix.An algorithm of mosaic is proposed different from existing methods.For each point in the point cloud,normal vector is calculated by searching the neighbor area,and the curvature value of each point can be obtained by means of coordinate transformation and parabolic fitting.An appropriate threshold is set to select the feature point set.A similarity measure is constructed to determine the corresponding points.Three paires of points for optimal registration are determined by using vector constraint and rotation transformation matrix and translation vector are calculated using these three paires of points and initial registration is completed.Similarily,the improved ICP algorithm is applied for precise registration.Experiments designed indicate that the algorithm has high accuracy.