Non-rigid 3D Projective Reconstruction Technique Based On Image Sequence

Because of the fixed structure and simple movement of the rigid object,the early research of 3D reconstruction mainly focuses on the rigid motion.However,most of the motion of objects in the real world belongs to non-rigid motion,and the dynamic non-rigid motion is more universal and diverse than the static rigid motion.Therefore,the traditional rigid 3D reconstruction method is hard to be applied in non-rigid 3D reconstruction.If we simply assume the non-rigid object as a rigid object,it can not reflect the objective world,and even it would lead to the failure of 3D non-rigid reconstruction due to the superposition error.Therefore,how to reconstruct the 3D structure of the non-rigid object from an image sequence rapidly becomes one of the hot issues in the computer vision community.In the absence of any prior knowledge,it can only achieve the non-rigid 3D projective reconstruction from an image sequence.Projective reconstruction is an essential part of non-rigid 3D reconstruction process,and the projective reconstruction precision will influence the final non-rigid 3D reconstruction effect.Therefore,in this paper,we focus on the research of non-rigid 3D projective reconstruction based on an image sequence.The main research work of this paper is as follows:(1)Images are stored in the form of a pixel matrix what is non negative matrix in the process of 3D reconstruction.Therefore,the non negative matrix factorization algorithm is applied to the non-rigid 3D reconstruction.We assume that the camera is a orthographic projection model,a non-rigid 3D projective reconstruction algorithm based on non negative matrix factorization is presented.The algorithm can ensure that the non negativity of the decomposition results and make the 3D reconstruction result more physical.The experiments with both simulate and real data show that the algorithm has high projective reconstruction accuracy.(2)In order to conform to the actual situation and improve the reconstruction accuracy,we assume that the camera is a pinhole model,an iteration non-rigid 3D projective reconstruction algorithm based on minimal eigenvalue is presented.All the image points and the depth factors constitute a low rank image matrix.Based on the characteristic of the low rank image matrix,the algorithm replaces projection solution with matrix eigenvalue and eigenvector solution.We can obtain the depth factors by iteration.Finally,we realize the non-rigid 3D projective reconstruction.The algorithm can guarantee to converge to the global optimal solution.The experiments with both simulate and real data show that the algorithm has the advantages of fast convergence speed and small error.