Navigating a Portable Robotic Device by a 3D imaging sensor

This paper presents a novel solution to the autonomy of a Portable Robotic Device (PRD) for the visually impaired. The proposed method meets the PRD's requirement of providing 3D navigational information with a small-sized device. The proposed approach is to employ a 3D imaging sensor-the SwissRanger SR4000-for both pose estimation and perception. The SR4000 produces both intensity and range images of its surrounding. The key idea for pose estimation is to perform feature extraction and matching between successive intensity images and use the 3D locations of the matched features to determine the PRD's pose. The method for obstacle detection/recognition is to segment the sensor's range data into planar surfaces, group the surfaces into structures, and identify the structures based on their geometric constituents. The proposed segmentation method is based on graph theory and it uses global features to measure inter-segment dissimilarity. The proposed methods' efficacies have been validated by experiments.