The Key Technology Research Of 3D Reconstruction Based On Kinect

The three-dimensional reconstruction(3D)is an important research direction in computer vision and graphics areas.The goal of 3D reconstruction is to establish 3D model of the object and scene,which can be used for computer's processing and analysis.This research achievement can be used for animation production,virtual reality,reverse engineering,robot navigation and so on.The kinect sensor produced by Microsoft can obtain the depth map conveniently,which pushing the 3D reconstruction to a new level.The three-dimensional reconstruction based on Kinect consists of four main stages: depth map acquisition,depth map conversion,point set registration,point cloud joint.The efficiency of point set registration affects the accuracy of 3D reconstruction directly.Point set registration is divided into rigid and non-rigid registration according to the registration property.Therefore,this paper does the deep-going research for rigid and non-rigid registration.The research content is as follows:1).The algorithm of depth image registration based on Spin-images and IRLS-ICP(Iteratively Re-weighed Least Squares Iterative Closest Point)is proposed.ICP(Iterative Closest Point)algorithm is the mainstream of the current rigid registration algorithm.However,the ICP algorithm is sensitive to initial correspondence.In order to solve this problem,we use the Spin-image coarsely to match the point sets firstly,which to find the roughly corresponding relations between them.Next,we make use of ICP algorithm to register the point clouds accurately on the basis of coarse matching.The experiments using the public data sets show that the proposed method can well be used for rigid registration.2).The algorithm of depth image registration based on IRLS-ICP-TPS(Iteratively Re-weighed Least Squares Iterative Closest Point Thin-Plate Splines)is presented.Rigid registration method cannot effectively register these point sets with deformation in three-dimensional reconstruction.So,this paper introduces TPS(Thin-Plate Splines)algorithm based on ICP method to solve this problem.Firstly,IRLS-ICP is used to find the control points between the point sets,which will be as the foundation of calculating non-rigid coefficients.Then,we use TPS to calculate the transformation matrix based on control points.Finally,the matrix is applied to one of the point sets to complete registration.The experimental results using the public data sets and homemade data show that the proposed method can greatly be used for non-rigid registration.