Abstract
The safety of the apron is crucial for aviation operations, and wingtip
scraping incidents are a common occurrence in this area. However,
existing airport apron conflict detection models have shortcomings such
as low accuracy and poor real-time performance. This paper proposes a
apron aircraft conflict detection model based on keypoint detection,
which detects potential wingtip accidents by fusing pixel and real-world
coordinate conversion methods and the improved YOLOLv8-Pose algorithm.
Results on the wingtip conflict detection dataset indicate that our
model, compared to the original YOLOv8-Pose, achieved an 87.3% increase
in FPS, a 96.76% reduction in the number of parameters, and an 82.14%
decrease in computational complexity.