A technique to segment by tracking extended targets manoeuvring through complex scenes

We present an innovative optical flow based algorithm that uses a memory of both target appearance and motion in order to simultaneously segment and track extended targets through complex scenes. A particularly attractive feature of this approach is that is assumes little prior knowledge of the scene content (background, clutter etc) and can cope with a variety of target types and numbers. The algorithm will be demonstrated on synthetic and real-world, visual-band imagery.