Target recognition and tracking method for one-off aerial imaging system

One-off aerial imaging system could work in rugged environments too hazardous for piloted or autonomous aircraft to cruise through, thus it has abroad potential utility in disaster assistance, exploration and some other fields. In this paper we develop a method for target recognition and tracking in image sequences from this kind of system. Target recognition is performed with local features of salient points, which could cope with partially obscured situation. We choose SIFT descriptor that is able to perform effective recognition with incoherent frames of one-off imaging system. Matching vector in high dimension is accelerated by our categorization scheme with a decision tree. In our method, no target detection operation is needed as targets are recognized directly by matching salient points. When initialization for tracking is finished, fast tracking methods start to record their position in following frames. The recognition algorithm is performed periodically to cope with entering and leaving problem and provide context for tracking method. Test with simulated image sequence demonstrates that our method can recognize and track predefined objects effectively in image sequences from one-off imaging system.