Estimation of Deformable Structure and Motion

Introduction The estimation of structure and motion from image sequences is one of the most studied problems within computer vision. However, almost all the efforts in this area have dealt with rigid objects. Our surrounding environment is generally not a rigid place, with swaying trees, moving people and rolling waves. Hence, to have structure and motion systems work effectively in our world, in general, the assumption of rigid objects has to be relaxed. We present an approach for estimating the structure and motion of deforming or non-rigid objects. This is done by employing the Principal Component Analysis (PCA) framework, whereby the object is supposed to deform according to a linear model. That is, the structure of frame i, S i = Q i1 Q in ; where Q ij denote the coordinates of a 3D point, can be expressed as S i = S + r k=1 ik S k ; (1) where ik is a scalar, S k is a 3D mode of variation and S is the mean shape. These types of models have proven to