Object Tracking Algorithm Based on Adaptive Learning Parameter by Online Loss Detection

摘要

The goal of this paper is to improve the robustness of the tracking by detection system. We gain the aim by dynamic tuning the learning parameters for an accumulative learned classifier according to online performance evaluation. Firstly, two kinds of tracking loss measures are distinguished: one is positive due to object appearance variation which does not result in drift; the other is negative which really causes drift. Then, according to these two situations, different parameter updating strategies are proposed to maintain adaptive model updating. Our proposed algorithm adopts the ability of identifying track loss and controls the tracking process to enhance the stability and robustness of the original algorithm (CT), and experimental results show better performance with negligibly small additional running time on some challenging video fragments which original CT failed on.