Dynamic Obstacle Detection using Temporal Hue Bounded-Disparity Maps with Artificial Neural Network Extended Depth

Sahan Fernando (PhD Student)

Dept. of Computer Science and Software Engineering, University of Canterbury

Wed Mar 05 10:15:00 NZDT 2008 in Room 315, Erskine Building

Abstract

This presentation presents a research on stereo vision, with three cameras for three different baseline lengths to identify dynamic obstacles within their range quickly and accurately. For obstacles beyond disparity level supervised learned ANN is proposed with novel disparity proximity map algorithm employed in hue image using bounding box algorithm instead of using another device to make it closer to human visual perception. Prior research attempts to detect dynamic obstacles using hybrid of stereo merged with other active sensors (such as laser scanners or millimetre wave radar, etc) because stereo alone has limited depth resolution being conversely proportional to the baseline length.