Monocular Image-based Intruder Direction Estimation at Closest Point of Approach

This paper deals with monocular image-based aircraft Sense and Avoid for small UAVs. After summarizing previous results of the authors it proposes a complete solution which calculates time to closest point of approach, relative closest point of approach (CPA) and the direction of intruder at CPA. These parameters are enough to make a collision decision and design the avoidance maneuver. The applicability of the proposed solution is demonstrated considering an omnidirectional multi-camera system in an extensive software-in-the-loop test campaign covering the whole possible size and velocity range of manned aircraft as intruder. Straight aircraft paths with constant velocity and camera pixelization errors were considered. Almost 100% decision success was achieved. After the simulations the solution is demonstrated on real flight test data and even in real flight giving 90% avoidance success (10% missed detection) in close and 60% decision success (40% false alarm) in far encounters.