Obstacle avoidance using time-to-contact information

Abstract

Machine vision can give rich information about the environment. However, image processing usually demands high computational effort. This paper proposes a framework for a vision-based row-data range sensor for obstacle detection using the concept of time-to-contact. Differently from usual approaches, here this quantity is computed directly, without involving a previous motion field estimation (via optic flow calculation, for example). Theoretical and practical comparisons between the direct approach and the optic-flow-based one for computing time-to-contact are presented. An experience that uses the proposed vision sensor to control a real robot was performed with good results.