Map-enhanced visual taxiway extraction for autonomous taxiing of UAVs

In this paper, a map-enhanced method is proposed for vision-based taxiway centreline extraction, which is a prerequisite of autonomous visual navigation systems for unmanned aerial vehicles. Comparing with other sensors, cameras are able to provide richer information. Consequently, vision based navigations have been intensively studied in the recent two decades and computer vision techniques are shown to be capable of dealing with various problems in applications. However, there are signi cant drawbacks associated with these computer vision techniques that the accuracy and robustness may not meet the required standard in some application scenarios. In this paper, a taxiway map is incorporated into the analysis as prior knowledge to improve on the vehicle localisation and vision based centreline extraction. We develop a map updating algorithm so that the traditional map is able to adapt to the dynamic environment via Bayesian learning. The developed method is illustrated using a simulation study.