Boundary Extraction of Moving Objects from Image Sequence

초록

In this paper we address an algorithm of contour extraction of moving object using Active Contours - so-called Snakes - in consecutive images. Both spatial convolution process and dynamic direction code are used to resolve the problem of automatically locating the initial geometric Active Contour for the external boundaries of an object in 2D image. Ongoing directions in which the snakes move are decided by three directions of eight directions. Owing to definition of specified directions, we could reduce the time complexity of energy minimizing operation of snakes by considering not eight but three directions. In order to applying the proposed algorithm to consecutive imagery, the Kalman filter, an optimal parameter estimator, is adopted. Our algorithm can be applied to extracting 2D information, which describes object, to recognize moving object in motion analysis system.