REAL-TIME COLLISION AVOIDANCE FOR UAVS USING MONOCULAR CAMERA

摘要

Successful operation of unmanned aerial vehicles (UAVs) in uncertain environments requires an ability to detect and avoid obstacles in the flight path ahead of the vehicle. We have developed a compound detection and avoidance strategy for static obstacles by using only a monocular camera onboard the UAV and focusing on real-time capability. In the strategy, a given image is segmented into multiple patches by using a fast marching method. The patches that have consistent shape and spatial coherence are registered as obstacles in the image plane. Data (e.g., shape, size, position, and color) for the identified obstacle are stored after the registration. We implement a reactive guidance strategy to steer the UAV around the obstacle. Furthermore, we propose a novel range estimation method using passive image sensing as a potential input to the reactive guidance strategy. We present collision avoidance demonstrations in simulation and field tests. The approach will be applied to complex urban environments.