Illumination invariant visual tracking via Kalman filtering and efficient feature matching

In this paper an efficient and accurate tracking algorithm is proposed. This algorithm uses a Kalman filtering approach and a linear dynamical model to predict the object position in next frame. Once the predicted position is obtained, a search region around it is defined and the best match for the object feature is found. The proposed feature is robust under illumination variations, so the algorithm performs well under severe illumination conditions. This claim is verified through extensive experiments on different datasets that the proposed method outperforms the state-of-the-art algorithms, especially under illumination variations.

Keywords:

tracking , dynamical models , Kalman filter , Retina filtering ,