A first step toward a temporal integration of motion parameters

The author presents the first results in the development of a recursive method for the temporal integration of motion parameters. The algorithm is based on a correlation scheme for optical flow estimation. The technique applied is suitable for online applications, because two frames at a time are needed. At any time one estimates the localization of the focus of expansion and stabilizes it using Kalman filtering. The knowledge of the focus position in previous time frames is used to reduce the search space of corresponding patches in the successive ones. For its recursive aspect, the method can be seen as a reflex able to detect the motion direction of a mobile robot. How the algorithm works on real image sequences is shown