Vision-based Runway Detection and Landing for Unmanned Aerial Vehicle Enhanced Autonomy
Kyriacos Tsapparellas, Nickolay Jelev, Jonathon Waters, Sabine Brunswicker, Lyudmila Mihaylova
Abstract
Introducing autonomy is a task of paramount importance and is currently investigated in many areas, especially for autonomous cars and Unmanned Aerial Vehicles (UAVs). Most UAVs are still remotely human-controlled. A necessity is to implement on-board solutions, able to work in all weather conditions and at any time. Hence, on this topic, we give an overview of recent advances for vision-based landing of UAVs. A thorough classification of the main recently developed methods is introduced with a discussion of their advantages and disadvantages. The paper presents a new solution for autonomous UAV vision-based landing, focusing on runway detection using a hybrid approach combining multi-image matching, SIFT and object tracking. The results are evaluated and validated using simulated images sampled with the X-Plane 11 flight simulator and real-world videos collected during automated flights performed by the ULTRA vehicle, one of the biggest UAVs in the UK [1]. The statistical analysis from the validation of the proposed approach shows a high level of accuracy around 94.89% in clear weather conditions and real-time computational performance.