A coarse-to-fine active contour tracking method based on target distribution

Abstract

When human recognizes an object, whole appearance is firstly observed, and then the details in each part are recognized. Following this process, we propose a coarse-to-fine active contour tracking method based on target's distribution. Firstly, a kind of new energy model based on target's distribution is used as the active contour energy model; using this energy model the curve can describe the target more exactly. Then, we improve the curve evolution for active contour. In our method, the tracking is divided into two coarse tracking and fine tracking. Firstly step, the contour moves to the target around quickly by global translation, this is the step of coarse tracking; after the global translation, the curve start to change the shape to fit the boundary of the target, this is the step of fine location. Experiments show that this coarse-to-fine active contour can track the target in complex scene effectively and accurately.