Image Missing Value Recovery Based On Non-rigid Trajectory Basis

The purpose of the 3D reconstruction is to recover the 3D structure and motion information from 2D image,which is one of the hot spots in the research of computer vision.The results of the study can not only make robots have the ability to capture the dynamic object,realizing the environment interaction,but also is now widely be applied to the man-machine interaction,visual monitoring,virtual reality,etc.But for the non-rigid motion situation,it's necessary to further study on the imaging model and feature point missing problem.Because trajectory basis is predefined,it can be reduce the solution of the unknown to improve the robustness of the algorithm.At present,the researches on the non-rigid usually regard it as a series of linear weighted combination of trajectory basis.Nowadays,the study on the non-rigid motion mostly bases on orthogonal projection model or weak perspective projection model,which both need to meet certain conditions.In addition,when an object moves,it's inevitable to be occluded,leading to the image feature points not be tracked,which is also be solved.This paper in the space of trajectory researches on the non-rigid structure from motion based on factorization method.Under the perspective model,this paper accomplishes the projective reconstruction on the non-rigid,and it's proposed that the non-rigid missing data can be recovered.The main work of this paper is as follows:(1)The present study on the non-rigid is mostly based on the weak perspective projection model or orthogonal projection model,ignoring the location information or the depth information.To overcome above disadvantages,this paper realizes non-rigid protective reconstruction in the space of trajectory,based on perspective projection model.In the process of implemention,using the properties of the column vector and the row vector in the projection matrix,it reduces the error and improves the accuracy.(2)When it moves,some of the non-rigid feature points can't be tracked.This paper puts forward to recovering missing data on the basis of the reconstruction.Because it exists two unknown in the image matrix,this paper uses the method that one unknown is initialized,then it takes turn alternating iteration.Because trajectory basis is orthogonal to each other,so it is independent of each other,less error accumulation.What's more,trajectory basis is predefined,which reduces the solution to the unknown and improves the robustness of the algorithm.The results of real experiments show good accuracy and effectiveness.