Landmark detection for vision-based navigation using multi-scale image techniques

This article concerns with visual based guidance of a mobile robot platform. A multi-scale image analysis followed by a filtering procedure using disparity information is presented for the detection of landmarks. Good preliminary results at the prototype stage demonstrate the usefulness of a feature extraction method for object recognition as a component of a robot navigation system. An extension of the procedure to applications with stereo is developed. This affords an evaluation of the object regions that generate the extracted features and the elimination of feature points that arise from regions of partial object occlusion