Fast Compressive Tracking With Improved Classifiers

Visual tracking in a video is a challenging problem in computer vision. The core component of object tracker based on tracking-by-detection framework is a discriminative classifier, tasked with distinguishing between the target and the surrounding environment. Fast compressive tracking algorithm is utilized to cope with real time tracking which trained a classifier to distinguish foreground and background, however, it does not take into account the influence of previous positive samples, when target occluded, it is easy lead to tracking fail or drifting problem. This paper proposed a new samples extracted method which take the previous positive samples into classifier training, two sub-classifier are trained and combined to a strong classifier which is used to distinguish target. Experimental results demonstrated the effectiveness of our method.