Motion segmentation scheme using multichannel optical flow estimator

In this paper we propose a segmentation scheme which utilizes motion information of optical flow to extract the region of moving objects and the background. Optical flow is represented as velocity field which is calculated from two or more images of a sequence. Segmentation of motion from a video sequence is expected in many applications of motion recognition, orientation estimate and so on. Each segmentation steps consists of two stages. First, the velocity of the scene is calculated by spatiotemporal gradient method with multichannel integration. In the second stage, the ratio of the magnitude of mean and covariance of the local velocity is used as a criterion for motion segmentation.