The Research Of Reconstructing3D Structure Of Non-rigid Object Based On Optimal Trajectory Space Finding

3D reconstruction is recovering the3D structure and motion from2D image sequence. Thisis a key point and a difficult problem in computer vision and pattern recognition. And thereconstruction of three dimensional structure from motion of non rigid object is more difficult.Factorization method is widely used because of its good stability and accuracy. However,applying the factorization method which is designed for rigid object in static scene directly into3D reconstruction of non rigid object in dynamic scene still exists much difficulty and problems.The majority of the non-rigid object factorization algorithm based on the main idea that thethree-dimensional structure of non-rigid object is seen as linear weighted combination of a seriesof shape basis. It shows that reconstruction results of factorization method depend on theselection of shape bases. However, shape base is unique and each shape base is not identical. It isobvious that dancing flags and movement of the human body cannot use the same shape base inthe3D reconstruction, which means shape base does not have universal applicability. However,this problem can be solved under trajectory space. When recovering3D structure from motion intrajectory space, the trajectory of feature points is viewed as a linear combination of a series ofpredefined trajectory bases. So it is important to select a suitable trajectory basis and areasonable number of it. Different trajectory basis will leads to different results of3Dreconstruction. What kind of trajectory basis and how many trajectory basis can reconstruct3Dstructure of non-rigid object accurately? The existing algorithms do not make deep study on theselection of trajectory basis. The estimation of trajectory base number is an important problem intrajectory space factorization algorithm. However, the existing algorithms select a suitable basenumber according to a large number of experimental results. There is no special research ontrajectory base number estimation.Aiming at above problems, the main works of this paper are as following:(1) The influence made by trajectory basis selection on recovering3D structure frommotion of non-rigid object is analyzed in this paper. There are lots of wavelet basis can be usedin3D reconstruction in trajectory space. Each basis has its own characteristics, and it will getdifferent effect in application. Different wavelet base should be used in different occasions, Sothe type of the trajectory basis take an important place in the algorithm of3D reconstruction. Asthis, the characteristic of some common wavelet basis will be discussed, two classical wavelet basis will be compared, and the influence of the selection of wavelet base on3D reconstructionresult of non rigid object will be analyzed in this paper.(2) A trajectory basis number selection method based on adaptive threshold is proposed inthis paper. In trajectory space, the estimation of trajectory basis number is very important tofactorization method of three dimensional reconstruction, and the existing research designs littlealgorithms for trajectory basis. If selecting too much trajectory basis, the algorithm will cost toomuch time; but if selecting too little trajectory basis, the precision of the algorithm can not beguaranteed. So how to select a suitable number of trajectory basis is very important in trajectoryspace. Since the trajectory of feature points is the direct reflection of deformation degree of nonrigid object, it is possible to estimate the chance of the difference of the corner of internal featurepoints according to the difference of trajectory corner, and then selecting the number oftrajectory is feasible. In order to make the algorithm more practical, this paper selects the bestthreshold value through calculating the variance between classes according to the amplitude ofthe difference of corner. Then the number of trajectory basis is finally gotten after cornerdifference matrix barbarization.