A Real-time Visual Reflex For Autonomous Navigation

Abstract

In this paper a well defined case of autonomous car driving in a typical traffic jam condition is addressed. Such a situation has been faced looking at the relation between the rate of change of the area projected on the image plane by a moving object and its time-to-crash (TTC). A visual sensor has been developed which is able to detect expansions or contractions of the area shape, without any explicit computation of the optical flow field. This sensor provides the visual input of an opto- motor reflex, working in real-time, which is able to keep constant in time the distance between the camera an a frontal obstacle. The methodology has been tested by using a mobile platform. The performances of the reflex on real image sequences are shown.