Landing is one of the most technically challenging phases of flight. Current precision approach systems based on ILS and GPS can greatly aid pilots but require ground infrastructure that isn’t available at every runway and can be vulnerable to signal interference or jamming. A purely vision-based approach offers a robust alternative that works at unequipped airfields, provides redundancy against spoofing or outages, and supports the growing push toward autonomous flight. Reliable monocular pose estimation could meaningfully improve landing safety for aircraft
In our capstone project, we investigate computer vision methods for estimating an aircraft’s 6-DOF pose (i.e. position and orientation) relative to a runway during final approach and landing. Using a single forward-facing camera, we proposed and began testing a pipeline for runway detection, feature extraction, and pose estimation. Our pipeline leverages YOLO-based methods for runway segmentation, then evaluates the efficacy of both learning-based and geometric methods for pose estimation.