Detecting negative obstacles (ditches, holes, wadis, and other depressions) is one of the most difficult problems in
perception for unmanned ground vehicle (UGV) off-road autonomous navigation. One reason for this is that the width
of the visible portion of a negative obstacle may only span a few pixels at the stopping distance for vehicle speeds UGV
programs aspire to operate at (up to 50kph). The problem can be further compounded when negative obstacles are
obscured by vegetation or when negative obstacles are embedded in undulating terrain. Because of the variety of
appearances of negative obstacles, a multi-cue detection approach is desired. In previous nighttime negative obstacle
detection work, we have described combining geometry based cues from stereo range data and a thermal signature based
cue from thermal infrared imagery. Thermal signature is a powerful cue during the night since the interiors of negative
obstacles generally remain warmer than surrounding terrain throughout the night. In this paper, we further couple the
thermal signature based cue and geometry based cues from stereo range data for nighttime negative obstacle detection.
Edge detection is used to generate closed contour candidate negative obstacle regions that are geometrically filtered to
determine if they lie within the ground plane. Cues for negative obstacles from thermal signature, geometry-based
analysis of range images, and geometry-based analysis of terrain maps are fused. The focus of this work is to increase
the range at which UGVs can reliably detect negative obstacles on cross-country terrain, thereby increasing the speed at
which UGVs can safely operate.