Research On Registration Algorithms For Scattered Point Cloud In Reverse Engineering

With the rapid development of digital technology and automotive industry,reverse engineering is increasingly popular in automotive manufacturing industry.In the forward design phase of the car,the treatment of the sludge model requires the support of reverse technology.For domestic automakers,there is still a big gap in the design level compared with foreign countries.The employment of reverse engineering to update the design of products is an important way to enhance competitiveness of automakers.In the process of data acquisition of reverse engineering,it is necessary to measure point cloud data many times due to the effect of various factors,such as object shape and acquisition equipment.In order to maintain the integrity of point cloud model,point clouds under different reference frames need to be transformed into the same coordinate system through point cloud registration.Aiming at the problems of low efficiency and accuracy of existing point cloud registration algorithms,this paper divides point cloud registration into two processes: coarse registration and fine registration,and studies them separately to improve the accuracy and speed of registration.The main work is described as follows:(1)Feature point extraction is performed by combining point cloud normal vector and density to complete point cloud coarse registration.Firstly,the point cloud normal vector is calculated by principal component analysis(PCA)method,and the point cloud density is calculated according to the distance between the point cloud data.The feature points are extracted by the characteristic degree corresponding to the point cloud normal vector and the detection parameters of the density component feature points.Then,the fast point feature histogram(FPFH)descriptor of feature points is calculated.Finally,based on the FPFH descriptor of the feature points,the random sampling consistency(RANSAC)algorithm is used to obtain the matching points,and the rigid transformation matrix is calculated by quaternion method to achieve coarse registration of the point cloud model.(2)An improved ICP algorithm is proposed to accomplish point cloud fine registration.Firstly,the search method of point cloud data is improved.The scattered point cloud data is indexed and ordered by a multi-layer index structure combining grid method and KD tree.Then,the normal vector is calculated based on the local surface fitting method.According to coordinate transformation method,the calculation of point cloud curvature is realized and the feature points are extracted.Finally,based on ICP algorithm,the corresponding point pairs are found in the curvature feature points,and the rotation matrix and translation vector are obtained by quaternion method to achieve fine registration of point clouds.The experimental results show that the proposed algorithm can achieve the fine registration faster and obtain smaller registration errors.(3)Taking the automobile model as an example,the algorithms proposed in this paper are experimentally verified and the whole process of reverse engineering combined with rapid prototyping is completed.Firstly,data collection is accomplished by RigelScan hand-held laser scanner,and point cloud filtering and hole repair are completed in Geomagic Studio.Then,coarse registration and fine registration of point clouds are conducted on MATLAB platform.Finally,the printing of automobile model based on Rapid Prototyping technology is realized,which verifies the effectiveness of the registration algorithm.