Online discriminative tracker with selective feature update

Visual tracking has been a challenge in the computer vision field for a long time now because of the many uncertainties present. Ideally, a tracker should be able to withstand certain problems in tracking with minimal computation time. Some of the difficult situations are background clutter, partial occlusions and illumination changes to name a few. In this paper, a discriminative tracker using SURF descriptors is improved by using a Particle Filter framework for translation prediction and a novel selective feature update for the classifiers. This implementation is to address the issue of fast motion and drift during the online-learning process.