Boundary Preserving Joint Estimation of Optical Flow and Disparity in a Sequence of Stereoscopic Images

The purpose of this study is to investigate a new method for the joint estimation of optical flow and disparity in a sequence of stereoscopic images. For a sequence of two pairs of stereoscopic images, the problem involves the estimation of four fields, two fields of image motion and two of disparity. Stereoscopy relates the four fields, and there is no relation between any three of these fields except through the fourth. We present a new method where three of the fields are estimated independently by anisotropic diffusion to preserve their boundaries, and the fourth deduced using the integrability constraint. Neither spatial nor temporal correspondence is explicitly referenced. Extention of the scheme over time requires the estimation of only two fields and an application of the integrability constraint at each time step. When motion or disparity in any of the fields is of large extent, the scheme is activated via multiresolution. Several experiments confirm the advantage of the method both in terms of accuracy and speed of execution.