IJCAS

Unstable object tracking usually happens in hand-held video sequences that consist of the movements of both tracked object and camera. The tracked objects in these sequences are highly unpredictable in displacement and appearance changes, and create more challenge than tracking with a static camera. In this paper, we propose a two-mode tracking model for dealing with unstable situations, such as scaling, pose changes, and abrupt movements. The short-range tracking mode is for the scaling and appearance changes. This mode incorporates Lukas and Kanade’s optical flow and the CAMShift, in which to achieve high accuracy, both color and corner point features are fused. The long range-tracking mode utilizes particle filter and CAMShift to capture fast and abrupt motion. We design a mode selection strategy based on a failure detection method for adaptation with each tracking case. The proposed tracking model shows high performance with difficult sequences against the recent tracking systems, as well as achieving real-time processing on smart phones.

Keywords: CAMShift, face tracking, optical flow, particle filter, smart phone.