A Context Based Robust Tracking Algorithm
Lead Presenter: Fei Xiong
Faculty Advisor/Principal Investigator: Octavia Camps
Method of Presentation: Poster
Tracking objects under occlusions is very challenging. Current approaches either rely on a Linear Time Invariant (LTI) model to estimate the target trajectory or only use the rigid image context to determine the target position. In this paper, an algorithm is proposed to predict the position of a hidden target by exploiting adaptive linear dynamics with respect to supporter features that can move with the target, independently of the target, or not at all. The proposed algorithm outperforms state of the art approaches by taking advantage of weakly associated context features to predict the target position without any prior assumption about the target dynamics, during long occlusions, and under severe camera motion.