Aerial sequence registration for vehicle detection

In this paper we present a system that aims at detecting and tracking moving objects from aerial video streams. These sequences are obtained from a camera mounted on an aerial vehicle which flies over roads and highways. In order to compensate the motion introduced by the dynamic behavior of the camera, we have to estimate geometrical transformations which stabilize consecutive frames. Our system is based on the use of Mutual Information as a criterion to find an affine model which relates a frame to a reference one. Then, moving regions are extracted using residual motion. The tracking module takes as input these regions to eliminate false alarms and to track vehicles all over the time.