Monocular camera based real-time on-board obstacle avoidance for unmanned aerial vehicles.

This work proposes a monocular based real-time on-board obstacle avoidance for UAVs. With observations from a single camera sensor, it builds a sparse 3D model of the visible part of the scene and applies a virtual force field to avoid colliding with any visible obstacle. This approach is based on visual SLAM systems, but it reduces substantially the computational requirements. It minimizes the amount of information stored and processed by focusing only on what exists in front of the drone instead of producing a full metric map of the environment.