Monocular optical flow has been widely used to detect obstacles in Micro AirVehicles (MAVs) during visual navigation. However, this approach requiressignificant movement, which reduces the efficiency of navigation and may evenintroduce risks in narrow spaces. In this paper, we introduce a novel setup ofself-supervised learning (SSL), in which optical flow cues serve as a scaffoldto learn the visual appearance of obstacles in the environment. We apply it toa landing task, in which initially 'surface roughness' is estimated from theoptical flow field in order to detect obstacles. Subsequently, a linearregression function is learned that maps appearance features represented bytexton distributions to the roughness estimate. After learning, the MAV candetect obstacles by just analyzing a still image. This allows the MAV to searchfor a landing spot without moving. We first demonstrate this principle to workwith offline tests involving images captured from an on-board camera, and thendemonstrate the principle in flight. Although surface roughness is a propertyof the entire flow field in the global image, the appearance learning evenallows for the pixel-wise segmentation of obstacles.
展开▼