next up previous contents
Next: Software Development Environment Up: Mobile Robotics Previous: Dynamic ReasoningNavigation

Visual Collision Detection for a Mobile Robot

Within the framework of active vision, we are concerned with the task of allowing an autonomous mobile robot to move freely about without collision in its environment, guided by a single video camera. We have developed a new framework for collision detection in which collision information can be directly inferred from a novel parametrization of the motion space. The parametrization procedure involves a prior velocity estimation stage to calculate normal components of velocity. In this stage, we also compute a reliability measure to evaluate the goodness of the velocity estimates. The velocity estimates are then combined using a voting technique which is robust to noise. A subsequent optimization step yields motion parameters from which we can determine both whether a collision will take place, and if so, when. Our approach is convergent with current models of visual motion processing in biological systems. Our two-stage algorithm, comprising a normal-velocity estimator followed by a motion integration stage, parallels what is known of motion analysis in, for instance, primate visual cortex, where earlier areas in the motion processing pathway feed primitive velocity information to higher levels of motion analysis. Moreover, the actual techniques we employ are inferred to be similar to what exists in visual cortex. Finally, our motion parameterization appears to dovetail with recent neurophysiological findings that the neurons in the medial superior temporal (MST) area of the macaque monkey are specialized for detecting different kinds of motion.
A. Lin, M.D. Levine

Thierry Baron
Mon Nov 13 10:43:02 EST 1995