Combining obstacle avoidance with robocentric localization in a reactive visual navigation task

This paper presents a novel approach to perform obstacle avoidance and robot localization using a single camera. This approach is based on the continuous detection and tracking of image features. Features are classified as ground points or obstacle points using the IPT (Inverse Perspective Transformation). Obstacle avoidance is achieved by means of a qualitative local occupancy grid built using the visually detected obstacle points, while the features classified as ground points are used to perform robocentric localization. The experiments, conducted indoors and outdoors, illustrate the range of scenarios where our proposal can be used, and show, both qualitatively and quantitatively, the benefits it provides.