A low-cost system using sparse vision for navigation in the urban environment

Abstract Established mobility aids, such as the long cane, enable visually impaired users to travel safely in urban environments. This paper describes continuing work to enhance this mobility by providing a series of high-level navigational goals. In particular, we describe algorithms to detect doorways and to facilitate centre-path travel. To maintain high performance at low cost, both algorithms use sparsely-sampled images from a single camera. Doorway detection is achieved by the detection of characteristic patterns of near-vertical and near-horizontal lines. The direction of travel along a path is determined by locating the dominant vanishing point of the lines in the image. Experimental results are presented for both algorithms.