An adaptive motion correspondence algorithm for multiple vehicle tracking from an air-borne platform

As high-quality remote images become available for civilian applications, the idea of gathering traffic data from remote sensing has been proposed. One key issue in this area is how to track the vehicles through the remote image sequences. This paper describes an adaptive motion correspondence algorithm for multiple vehicle tracking from air-borne remote cameras. The motion correspondence is established by searching over an extended correspondence matrix. An on-line confidence factor is assigned to each track to control the search strategy. Experiments with real traffic scenarios showed that the algorithm could automatically distinguish track initiation from false detection, as well as track termination from occlusion, which demonstrated its feasibility for real-time applications.