Indoor Robot Navigation with Single Camera Vision

Abstract

In the large area of autonomous robot navigation, encompassing map creation, path planning, and self-localization, we develop the idea of a simple autonomous agent relying only on vision information. Our integrated navigation system replicates some functions of natural systems, as using little a-priori knowledge, on board computation, no omni-directional vision. Since our goal is essentially to move the robot on a floor, avoiding obstacles and people, the camera is on top of the robot and in fixed orientation to look ahead at the floor. After a simplified calibration, floor images are taken and positions of obstacles detected New images dynamically grow a grid map, constructed with simple mathematics and heuristics. For path planning obstacles are enlarged in minimum way, and path computed.