Abstract

Following Bulthoff et al.,¹ we have designed a scheme that computes an optical flow field using independent measures of motion. The independent motion measurements are computed in parallel in different spatial frequency channels for all possible displacements. An area-based winner-take-all scheme leads to different voting distributions in each channel. These partial estimates of motion are integrated into a more robust optical flow field using an additional voting step through the channels. Provided an adequate architecture, the implementation of this scheme would retain the highly parallel characteristic of the original voting algorithm (Bulthoff et al., 1989). Results obtained from sequences of synthetic images indicate that the additional voting step compensates for false motion estimates in the independent channels. Our scheme shows how a better approximation of the optical flow could emerge from partial estimates. It is also physiologically plausible because it accounts for the fact that motion tuned cells in cortical area V1 are also spatial frequency tuned.

© 1989 Optical Society of America