An Optical Flow Based Electro-Optical See-and-Avoid System for UAVs

This paper presents a study on a fully electro-optical (EO) autonomous target detection system for UAVs. This study was led in the framework of a partnership between the Italian Aerospace Research Center (CIRA) and the University of Naples, within a CIRA project named TECVOL, aimed at developing the technologies needed to support flight autonomy. The selected detection algorithm is based on the Optical Flow technique. In particular, two methods are implemented and tested on real video sequences of flying aircrafts, which are the Horn and Schunck's and the Lucas and Kanade's algorithms. Target detection is based on motion field evaluation, so that pixels with a statistically different velocity from the background are recognized as objects. Numerical results show that the LK technique guarantees better performances, and it is more suitable for real-time implementation. In fact, in the case of fixed background, no miss-detections (MD) and false-detections (FD) can be noticed. Imposing a motion of the optical axis in order to simulate camera wobbling effects, again zero MD are measured, while on average 4,4 FD/frame are observed. However, detection performance gets worse for increasing optical axis oscillation.