Real-time Vision System for UAV Runway Detection
Author Information
Author(s): Tsapparellas Kyriacos, Jelev Nickolay, Waters Jonathon, Shrikhande Aditya M., Brunswicker Sabine, Mihaylova Lyudmila S.
Primary Institution: University of Sheffield
Hypothesis
Can a vision-led system improve the accuracy and efficiency of runway detection for UAVs during landing?
Conclusion
The proposed vision-led system can localise runway sidelines with an accuracy of approximately 84.4%, outperforming existing methods.
Supporting Evidence
- The system demonstrated high accuracy in various weather conditions, including fog and crosswind.
- Real-time performance was achieved with a processing time of approximately 0.007 s per frame.
- The optimal number of reference images for high accuracy was found to be between 10 and 15.
- Structured Forests edge detection outperformed the Canny edge detector in accuracy.
- The system can be applied to other UAV types beyond the fixed-wing model tested.
Takeaway
This study shows how a special camera system can help drones land safely by finding the edges of runways, even in tricky weather.
Methodology
The study used a combination of computer vision techniques, including SURF for feature extraction, Structured Forests for edge detection, and Kalman filtering for predicting runway sidelines.
Limitations
The system was tested primarily on a fixed-wing UAV and may require adjustments for other aircraft types.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website