Monocular-Vision-Based Precise Runway Detection Applied to State Estimation for Carrier-Based UAV Landing

Area of Science:

• Robotics
• Computer Vision
• Aerospace Engineering

Background:

• Enhancing autonomy in the landing phase is crucial for full-envelope autonomous flight in unmanned aerial vehicles (UAVs).
• Accurate runway detection and precise UAV state estimation are critical challenges for autonomous carrier-based UAV landings.

Purpose of the Study:

• To propose an end-to-end state estimation method for autonomous landing of carrier-based UAVs using monocular vision.
• To improve the accuracy of runway detection and the precision of UAV state estimation during the landing phase.

Main Methods:

• A robust runway detection framework based on YOLOv5 (you only look once) with four modules: data augmentation, feature extraction, feature aggregation, and target prediction.
• Integration of a corner prediction method utilizing geometric features to enhance landing field prediction accuracy.
• Development of monocular vision-based datasets for carrier-based UAV landing simulations and implementation using the PyTorch deep learning tool.

Main Results:

• The proposed method demonstrated higher precision in runway detection.
• Improved performance in UAV state estimation during autonomous landings.
• The YOLOv5-based framework with geometric corner prediction effectively identified landing sites and estimated UAV states.

Conclusions:

• The developed end-to-end state estimation method enhances autonomous landing capabilities for carrier-based UAVs.
• The approach successfully addresses the accuracy requirements for runway detection and state estimation.
• This vision-based system facilitates safer normal and emergency landings by enabling in-flight landing site discovery.