Abstract:
Landing is the most difficult phase of flight for a drone. Due to lack of
efficient systems, there has been numerous accidents resulting in damaged
hardware. Vision based systems have been proved significantly popular in
correct detection of landing site, mainly because of their low cost and less
equipment involved.
This research focuses on detection of runway in aerial images with untidy
terrains as it can help a UAV to detect landing targets that is runway, and
thus helping in automatic landing. The scope of this research is accurate
detection and localization of runway in aerial images. Most of the work re garding runway detection is based on simple image processing algorithms
with lot of assumptions about position of runway in the image.
First part of this research is runway detection using deep learning and second
part is runway localization using both deep learning and non-deep learning
methods. In first part, land has been classified to detect runway in the image.
In second part, runway has been localized using line detection algorithms and
a CNN model. Land has been classified with accuracy up to 97% and runway
has been localized with IOU of 0.8.
