Vision-Based Sense and Avoid with Monocular Vision and Real-Time Object Detection for UAVs

Abstract

The use of unmanned aerial vehicles (UAVs) or drones have become ubiquitous in the recent years. Collision avoidance is a critical component of path planning, allowing multi-agent networks of cooperative UAVs to work together towards common objectives while avoiding each other. We implemented, integrated and evaluated the effectiveness of using a low cost, wide angle monocular camera with real-time computer vision algorithms to detect and track other UAVs in local airspace and perform collision avoidance in the event of a communications degradation or the presence of non-cooperative adversaries, through experimental flight tests where the UAVs were set on collision courses.