Autonomous track and land a MAV using a modified tracking-learning-detection framework

In our previous work, we mounted two separate sets of Pan/Tilt Unit (PTU) integrated with visible light camera on both sides of the runway for landing a Micro Aerial Vehicle (MAV) automatically. In this study, we focus on improving the precision of MAV tracking during the landing procedure. We seek to remedy the tracking-learning-detection (TLD) framework by using adapted Random Ferns methods and modified binary code system. Then, by introducing Extend Kalman Filter (EKF) to our framework, we make the algorithm more suitable for fully autonomous landing. Finally, several real flights in outdoor experiments show that the modified TLD has a better performance compared with our previous methods. It indicates that our approach can meet the requirements of robustness and real-time capability.