An Improved Object Tracking Method in UAV Videos

Abstract Unmanned aerial vehicles equipped with surveillance system have begun to play an increasingly important role in recent years, which has provided a wealth of valuable information for national security and defense system. The automatic understanding technology based on video contents becomes especially important when facing so abundant information. The research of objects tracking in understanding video contents is necessary. In many tracking algorithms, Mean Shift algorithm has been standing out as its efficient pattern matching and fast convergence, it can track objects in real time if the targets’ initial areas are known and it is robust to target sheltering and distortion, however, the lack of necessary update algorithm to modify the objects’ template in time, Mean Shift algorithm cannot adaptively track objects when the size of a moving object changes. At the same time, Mean Shift algorithm cannot track fast moving objects. Our thesis utilizes particle filtering based Mean Shift algorithm to optimize the searching origin of Mean Shift algorithm, and it accomplishes tracking fast moving objects. Besides, compared with Mean Shift algorithm, our algorithm can also adaptive to the size variation of moving objects.