Automatic Vehicle Tracking in Low Frame Rate Aerial Image Sequences

Traffic monitoring requires mobile and flexible systems that are able to extract densely sampled spatial and temporal traffic data in large areas in near-real time. Video-based systems mounted on aerial platforms meet these requirements, however, at the expense of a limited field of view. To overcome this limitation of video cameras, we developed a concept for automatic derivation of traffic flow data which is designed for commercial medium format cameras with a resolution of 25-40 cm and a rather low frame rate of only 1-3 Hz. Since resolution and frame rate are the most limiting factors, the focus of the first implementations and evaluations lies on the approach for automatic tracking of vehicles in image sequences of such type in near real-time. The tracking procedure relies on two basic components: a simple motion model to predict possible locations of previously detected vehicles in the succeeding images and an adaptive shape-based matching algorithm in order to match, i.e. recognize, the detected vehicles in the other images. To incorporate internal evaluations and consistency checks on which the decision of a correct track can be based, the matching is done over image triplets. The evaluation of the results shows the applicability and the potentials of this approach.