Visual Tracking Based On Matching Cascade

Abstract

With the increasing application of multi-target tracking technique, improving the tracking efficiency and processing of online data has become a hot issue. To solve the online multi -target tracking problem, this paper presents a hybrid data association method based on the comparison of local and global da ta associations. The method can guide global association with local constraints and seek global optimization for local associations. Objects and possible associations in video frames are thus abstracted. By constructing a cost function and calculating the lowest cost, optimal data correlation can be sought out and the optimal trajectory is subsequently acquired. Hybrid data association is then implemented on the real video frames which are chosen as the data sets for the tracking experiment in this paper. The performance evaluation is carried out and is compared wit h the existing multi-target tracking technology. The experiment result shows that the method performs well in many challenging environments and tracking is effectively improved.