A vision-based sense-and-avoid system tested on a ScanEagle UAV

This paper presents a study of near collision course engagements between a Cessna 172R aircraft and a ScanEagle UAV carrying a custom built vision-based sense-andavoid system. Vision-based systems are an attractive solution for the sense-and-avoid problem because of size, weight and power considerations. We present post flight test analysis that shows our detection system successfully detecting an approaching Cessna aircraft in all 15 flight test encounters at ranges greater than 1500 m, with no false alarms events. Moreover, this paper characterises the image inter-frame stabilisation required to achieve acceptable detection performance, and compares a range of stabilisation techniques for achieving this type of stabilisation precision. Our analysis illustrates that the image inter-frame stabilisation requirements are demanding, suggesting that images must be stabilised in real-time at 9Hz to within 2 pixels between consecutive frames. We present performance comparisons between stabilisation using GPS/INS, IMU-only and image-based techniques.