The system we describe here provides part of an obstacle avoidance capability for ASMONC (Autonomous System for Mobility, Orientation, Navigation and Communication), a project which aims to provide a full navigation and mobility capability for partially sighted people. A major requirement for the vision system in ASMONC is to detect small obstacles as well as kerbs/steps to help the user navigate safely along a path.
Ground Plane Obstacle Detection (GPOD) using stereo disparity was first reported by Sandini et al. [ 4 ] and subsequently refined by Mayhew et al. [ 7 ] and by Li [ 5 , 6 ]. It has been adapted to detect small obstacles for the partially sighted [ 11 , 8 ]. The Sobel edge detector used currently in GPOD detects vertical and near vertical edges, which are then matched by the PMF algorithm [ 9 , 10 ]. This works well in practice for obstacle detection as they are often standing on the ground.
However, apart from obstacles, a partially sighted person also needs to be informed of the presence and locations of kerbs or steps. We have developed a separate module using a general Canny edge detector [ 2 ] and the Hough Transform [ 12 ] to detect such obstacles.
Stephen Se