Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
411317 | Robotics and Autonomous Systems | 2014 | 17 Pages |
•The aim of this paper is to bridge the gap between monocular vision-based terrain classification and object detection in computer vision, by presenting a broad and structured overview of recent computer vision techniques behind the successes of object detection.•We believe that these techniques provide great potentials for terrain classification using monocular vision processing.
Direct terrain classification from monocular images for autonomous navigation of planetary rovers is a relatively new and challenging research area, not only because of the hardware limitation of a rover, but also because the rocks and obstacles to be detected exhibit diverse morphologies and have no uniform properties to distinguish them from background soil. We present a survey of recently developed object detection techniques that can be useful for terrain classification for planetary rovers. We start with summarizing current vision-based terrain classification methods. We then provide a comprehensive and structured overview of recent object detection techniques, focusing on those applicable to terrain classification.