Real-time object tracking via CamShift-based robust framework

摘要

In recent years, lots of object tracking methods have been presented for better tracking accuracies. However, few of them can be applied to the real-time applications due to high computational cost. Aiming at achieving better realtime tracking performance, we propose an adaptive robust framework for object tracking based on the CamShift approach, which is notable for its simplicity and high processing efficiency. An adaptive local search method is presented to search for the best object candidate to avoid that the CamShift tracker gets confused by the surrounding background and erroneously incorporates it into the object region. A Kalman filter is also incorporated into our framework for prediction of the object's movement, so as to reduce the search effort and possible tracking failure caused by fast object motion. The experimental results demonstrate that the proposed tracking framework is robust and computationally effective.