An active vision system for navigation in a natural environment

Abstract

The fusion of an active stereo camera system together with a mobile robot having the capability of orienting and navigating in a natural human environment is presented. This performance is achieved by realizing a task-specific hierarchy of visual, biologically motivated information processing modules. The claimed behavioral complexity follows on the one hand from the task-specific combination of simple, but versatile, units, and on the other hand from the active vision concept. This concept is able to solve basic vision problems in a much more efficient way than a passive one. The basic ideas of how such a hierarchy of capabilities can be achieved are described. The structure of the utilized mechanical system and the software concept founded on it are presented. Two basic modules in the field of low-level tasks are illustrated: a stereo obstacle detection algorithm and a motion detection scheme based on correlation. They function as examples of actually developed `evolvable' modules combining biological principles in order to realize a basic task in visually guided behavior