Abstract and Animation of M.S. Thesis

Object Tracking Algorithm using Deformed Template and Level-Set Theory

Hyun Tae Na
Department of Computer Science and Eng.
Graduate School of Hanyang Univ.
Supervised by Prof. Young Shik Moon

    Since object tracking is a basic and essential step for many video processing applications, many algorithms for tracking the accurate shape of an object have been proposed.
    In this thesis, an object tracking algorithm is proposed, which is robust against background variation, object flexibility, and occlusion. Toward this end, a template consisting of the object region and the deformation parameter is used. The parameters are updated at every frame, and the target frame is compared with the deformed template updated by deformation parameter. This process guarantees a tracking result which is robust against translation, rotation and scaling of objects.
    The proposed algorithm consists of two steps. The first one is an estimation of object shape and location, on the assumption that the transformation of object can be approximately modeled by the affine transform. The second step is a refinement of the object shape to accurately fit to the real object, by using the potential map and the gradient map.
    The proposed algorithm can track an object by using only the boundary information of 3D object projected onto D space. The experimental results have shown that it can also track a flexible object with large deformation.

Click images to watch animation.