On the use of motion as a primitive quantity for autonomous vehicle guidance

The goal of this work is to extend the navigational capabilities of an existing autonomous vehicle so that it can reliably traverse a real-time dynamic environment. While there is a wealth of literature on the extraction of motion information from a sequence of images, relatively few investigators have considered the problem of using this information to avoid dynamic obstacles. In this paper, we give a conceptual overview of this work. First, a system capable of extracting motion information in real time, using multiple sensors for increased robustness is introduced. Then our real time dynamic obstacle reasoning formalism is presented, showing how high level motion information can be utilised to improve existing navigation processes.