The Technology Research Of Point Set Registration Based On Depth Map

The point set registration is a significant research target in computer vision and image and graphics.The purpose of point set registration is to acquire the data of the scene or object that are used to be analyzed by the computer and get the 3D model of them.At present,the point set registration has a very wide range of applications,for example,virtual 3D map,robot navigation and 3D model printing in the entity shop.The point set registration based on the depth map usually contains three steps: depth map acquisition,depth image transformation,point cloud registration.Among them,the most important step is the point cloud registration.The point cloud registration accuracy directly affect the effect of 3 d reconstruction.Point set registration can be divided into rigid registration and non-rigid registration according to the target object whether producing deformation.ICP(Iterative Closest Point)algorithm is the best registration algorithm,which is used to rigid registration.The registration effect is more accurate through constant iterative step.Compared with the traditional registration algorithm,its accuracy is higher.For the ICP algorithm,the first important step is to find the initial corresponding relations between the two point sets.The good initial corresponding relationship can not only reduce the number of iteration but also reduce the running time of the program.Sometimes,the object will produce deformation,the ICP algorithm can’t effectively registrate the deformation of the point cloud now.To solve the above problems,this paper main research the registration algorithm;The main research content is as follows:(1)The algorithm of point sets registration based on rough matching and Sparse ICP is proposed.ICP algorithm is strict with the initial point sets corresponding relationship.In order to solve this problem,firstly,using the coarse matching to registrate the two point sets,finding the roughly corresponding relation between them,and then using the roughly corresponding relation as the initial correspondence of Sparse ICP.Finally through continuous iteration,until the average distance between two points is less than the threshold or reaching the maximum number of iteration of preset.Then the best of registration is produced.Through the experiments on the datasets show that the method can be used for rigid registration.(2)The algorithm of point sets registration based on Sparse ICP and TPS(Thin-Plate Splines)is presented.Sometimes,the object will produce deformation.ICP algorithm is rigid registration algorithm which is not suitable for deformation of non-rigid registration.Based on this problem,in this paper,after finding the accurate corresponding relationship using ICP algorithm,we choose some points as the control points which will be used to calculate the non-rigid coefficients between the points.Then we will find a transformation matrix.Through the experiment and analysis show that the method can be used for registration of deformation object.(3)Realizing the display of data set and triangular grid of 3D model of.mat files.The display of point set sometimes can’t express the whole complete model;And triangulation method can show the 3D model is more complete.