Efficient Surf Tracking by Nearest Neighbor Classifier

Abstract

Visual object tracking is a fundamental problem in many real applications, like video security surveillance, human computer interaction (HCI), video communication and compression, augmented reality, traffic control, medical imaging and video editing. The common challenge towards this task is the ambiguity existing among object and the background. To differentiate object from background, SURF descriptor [12] is applied in this paper to enable the efficient tracking-by-detection. SURF descriptors are extracted only in region of interest in each frame, to ensure the tracker’s high efficiency. Under the recently proposed framework, nearest-neighbor criterion is used to match the corresponding feature points within or out of the object appearance. Efficient sub-window search method is applied to accelerate the searching process while without sacrificing to searching accuracy. Experimental results prove the proposed tracker could be used in a real time tracking scene.