Tracking the rotating targets in aerial videos

In aerial videos, there are two main situations where the appearance of a target would vary. One is occlusion, and the other is rotation of the target. The former could be solved by track forecast which may use Kalman filter, Particle filter or some other similar methods. Aimed at the latter situation, incremental visual tracking (IVT) has been brought up. However, in aerial videos the rotation of targets (vehicles mainly) mostly occurs in the imaging planes, where IVT cannot make the bounding box rotating the same rotation angle of targets. Accordingly, we present a tracking system with specially designed methods that solves the issue. And it could track the rotating target continuously and give out the rotation angle in aerial videos. With a rotatable bounding box to indicate the target, the observers would read the information from the screen more easily and the system would extract the target status more accurately. The system is developed employing Correlation analysis and Kalman filter as well. Experimental results are presented on several aerial video sequences captured by the authors.