Adaptive Visual Tracking via Learning Detector of Specific Landmarks

Abstract

It is known that visual tracking is important in many applications. One of its difficulties is the track of fast-moving object in random motion, especially in the field of robot vision. In this paper, under the challenging conditions (e.g., complete occlusion and random movement) a novel Adaptive Visual Tracking via Learning Detector of Specific Landmarks (AVTLDSLs) is developed to predict the location of object (i.e., landmark or target). The problem of long-term visual tracking of unknown object in unconstrained environments is robustly tackled by the proposed AVTLDSLs. The experimental results of challenging videos and the comparisons between our AVTLDSLs and other method are presented to evaluate the superior accuracy and robustness of the proposed method.