Abstract: The paper presents a new method for optical flow estimating, which combines feature based with flow based method. By using the corner points as feature points and estimating the optical flow from image sequence, the optical flow is estimated by measuring the displacement of sparse located corner points between consecutive frames. It is proved theoretically that optical flow can be replaced approximately by using the displacement field. The paper also estimates 3D motion and structure from optical flow in image sequence. In this paper, the implementation of a non-linear algorithm is described,whose uniform observability, minimal realization and stability are proved analytically. Experimental results show that the new method provide a good estimation of the optical flow and 3D motion and structure.

Key words: computer vision, optical flow, corner point displacement field, extended Kalman filter, 3D motion and structure, image sequence