Research On Rigid And Non-rigid Point Set Registration Based OnIterative Linear Optimization

In recent years,point set registration has become a research focus in the field of computer vision,medical image analysis and processing,pattern recognition and so on.Point set registration technique has important application in the field of motion tracking,shape matching,image retrieval,image matching and so on.With the rapid development of computer aided technology,people put forward higher requirements in the aspect of registration speed,registration accuracy,registration rate,robustness of point set registration algorithm and so on.Aimed to traditional point cloud registration algorithm one or more problems of slow of the registration speed,large of the residuals,low of the registration,poor of the robustness and so on,this topic research on 2D/3D rigid and non-rigid point set registration algorithm.Firstly,basic thoughts of traditional point cloud registration algorithm theory were investigated,shortcomings and corresponding reasons were analyzed,laying the foundation for improving registration algorithm.Secondly,by analyzing the problem of traditional point set registration algorithm,this paper presents a novel point set registration algorithm based on iterative linear optimization,which can be used to register both rigid and non-rigid point set.A new cost function was constructed to evaluate the summation of squared distance between the two point sets,in which rigid transformation,non-rigid elastic deformation and complex deformation were all included for consideration.Thus,this paper were soundly solved the requirements which can't meet the fast registration,high precision,strong ability to resist abnormal deformation at the same time.Finally,by contrast with the robust point matching algorithm based on thin plate spline and registration algorithm based on single dynamic model,the performance of this paper algorithm is demonstrated and validated in a series of experiments of 2D point set registration,3D point set registration and registration with noise points in initial point set.