Outdoor RGB-D SLAM Performance in Slow Mine Detection

Abstract:
The introduction of Kinect-style RGB-D cameras has dramatically revolutionized robotics research in a very short span. However, despite their low cost and excellent indoor performance in comparison with other competing technologies, they face some basic limitations which prevent their use in outdoor environments. Perhaps for the first time, we report an outdoor application of Kinect-style cameras in a robotics application. We report the use of benchmark RGB-D SLAM algorithm in a land-mine detection application. We argue that due to the slow nature of mine-detection tasks, basic limitations of RGB-D outdoor performance are overcome or gracefully negotiated. We report results from extensive experimentation in a variety of terrains & lighting conditions for three different types of robots.