Liang KH & Tjahjadi T

Applied Signal Processing, vol. 4 no. 3, 1997 (printed in 1998), 179-187

Abstract

The segmentation of optical flow fields is an important stage in real-time object tracking. The multi-point optical flow approach is a means to achieve this, but the approach is limited by uncertainty. This paper presents a segmentation scheme which incorporates multiresolution classification and the multi-point method to overcome the uncertainty. The classifier involves a spatio-featural focusing and a fixing process. The segmentation scheme produces a multiresolution segmentation map, where the central region of a physical object is represented using coarse resolutions so as to achieve a better estimation of the optical flow, and the border region is represented using fine resolutions so as to achieve a better estimation of the object boundaries.