| In general, motion analysis involves the motion detection and measurement, and interpretation. In higher animals, this interpretation gives rich information about the world registered in the visual system, such as 3D shape of the object, segmentation, and disparity information, etc. It appears that biological systems are capable of doing this efficiently and accurately. It is for this reason that researchers have attempted to find a systemic architecture for motion analysis in a biological system. Recent studies have found many interesting organizational features for the biological visual system.; On the other hand, the computer vision systems have difficulty in simulating even simple tasks such as global motion measurement. This rather poor performance is due to many reasons. For example, the optical flow method is inherently ill-constraint problem which requires an additional constraint or assumption. In many cases, these assumptions are invalid. The estimate of optical flow is often inaccurate at the motion boundaries where the characteristics change discontinuously. It also requires the integration of local velocity field to solve the aperture problem. The matching techniques also have their own problems such as the amount of preprocessing required, what to match, how to match.; In this study, we propose a computational architecture for the detection and segmentation of coherent motion using multilayer cooperation/competition scheme. We apply knowledge obtained from the psychological/psychophysiological studies to build the system. Using the proposed system, we show how ambiguous local velocities are fine-tuned to recover unambiguous local velocities and how these recovered local velocities can be segmented using oscillatory correlation. |
