Fast boundary extraction for mobile robot sensor-detected images

A fast boundary/line extraction algorithm used for grid-based world maps in the dynamic navigation of a mobile robot is described. The usual maps for such devices are generated from ultrasonic sensor information and are stored in large pixel arrays. The utilization of these maps often involves the processing of every pixel in the map array, which necessitates relatively long computation times. The authors have transformed the image pixels of critical object boundaries into straight-line segments in order to reduce the computation time for tasks that depend on the map, such as path planning and map transmission. The image array is scanned only once by the algorithm while a carefully selected operator (mask) set is convolved with the image. At the same time candidate line-support pixels are sorted into eight lists. The object boundary pixels are grouped and converted into lines that require only two endpoints. Test runs for an experimental obstacle field indicate that this algorithm requires less than 1 s for line extraction, which is fast enough for real-time navigation