Researches On Object Tracking In Aerial Video Sequences

Tracking objects and further inferring its behaviors in aerial video sequences have wide applications in both civil and military domains such as urban surveillance, traffic control, battle field reconnaissance, homeland security, etc. Full or partial automation means drastically diminished man laboring hours and casualties.This paper concentrates on building a system capable of tracking objects through aerial video sequences. In the paragraphs, we point out the nature of motion compensation equals to calculate inter-frame geometry transformation matrix. We introduce the SIFT (scale-invariant feature transform) technique into the motion compensation module which yields extremely accurate results. Matching criterion is also of importance to a tracking system, we first noticed the limit of using Bhattacharyya distance as a matching gauge. This paper culminates in a matching criterion that we proposed based on statistical appearance of objects. We insert our criterion into the framework of particle filter where the former serves as the observation likelihood function.Experimental results show that our tracking system has the ability to disambiguate genuine objects from background and other moving things with similar color and that our system, under very shaky conditions, still can hold to its objects.