Research On Point Cloud Registration Method Based On 3D Camer

With the continuous development of computer vi sion technology,3D point cloud are gradually becoming a way to depict the real world.As an important part of 3D point cloud processing,point cloud registration plays an important role in3 D reconstruction,reverse engineering,medical image and heritage conservation.In order to solve the problems of low accuracy and poor robustness in point cloud registration,this paper investigates the point cloud registration methods from two aspects,coarse registration and fine registration,respectively,and the main research contents are as follows:(1)An improved SAC-IA coarse registration algorithm(KD-SAC-IA)based on key point is proposed to address the problems of time-consuming computation and low accuracy of registration when the number of point clouds is large.The key points of point cloud are extracted by three features: curvature,normal angle and density,and the distance constraint is introduced in the SAC-IA registration process to eliminate the set of incorrectly matched points,the rigid body transformation matrix is calculated based on the remaining set of points to complete the coarse registration.For the point cloud with low overlap rate,the coarse registration is performed by combining with the rotary table.The experimental results show that compared with some existing algorithms,the registration accuracy of the KD-SAC-IA algorithm proposed in this paper is improved by about 31% and the registration efficiency is improved by about78%.(2)To address the problems of high initial position requirement and easy to fall into local optimal solutions that occur in ICP algorithms,an ICP algorithm that incorporates a two-way search tree is proposed(UD-ICP).The point clouds to be registered are uniformly downsampled,then K-D trees are built for the downsampled point clouds separately,and the corresponding point sets are determined by mutual search,the rigid body transformation matrix is calculated according to the corresponding point sets to complete the fine registration.The experimental results show that compared with the ICP algorithm and some ICP derivative algorithms,the UD-ICP algorithm proposed in this paper improves the registration accuracy by about 20% and the registration efficiency by about 60%.(3)A multi-frame point cloud continuous registration strategy is proposed to address the redundancy and error accumulation problems in the continuous registration process,adaptive downsampled is used to process the results of each registration,the final results is optimized by the moving least squares method.The experimental results show that the continuous registration strategy and the global optimization method proposed in this paper can successfully perform continuous registration of multiple point clouds and solve the problem of error accumulation.(4)A point cloud registration system is designed and implemented.The system is developed in conjunction with the SDK of 3D camera,and mainly contains three components: data acquisition platform,software platform and application algorithm,which can realize point cloud acquisition,point cloud filtering,single registration and continuous registration functions.The above results show that the point cloud registration methods proposed in this paper can obtain good results in both public data sets and data collected by3 D camera,and meet the registration requirements in practical applications.