A robust approach for FOD detection on runway powered by image deblurring

Abstract Detecting Foreign Object Debris (FOD) on runways is crucial for ensuring flight safety and preventing accidents. However, this task becomes significantly more challenging in adverse weather conditions like rain, snow, and fog. This underscores the need for a robust object detection system capable of functioning effectively in blurred environments. In this study, we propose Deblur model, a dedicated convolutional neural network (CNN) model designed for deblurring tasks. Unique aspect of this study is integration
of Deblur model with the You Only Look Once (YOLOv8) object detection architecture. Deblur model aims to enhance FOD detection on runways under blurred conditions. A notable divergence from conventional object detection algorithms that frequently falter in low-visibility situations. Deblur model substantially elevates the quality of input data fed into the YOLOv8 detector. This enhancement in input fidelity directly translates to superior detection accuracy and robustness. This integration is a novel method of addressing the two problems of object detection and deblurring in one framework. We evaluate the effectiveness of this integration through a comparative analysis on a blurred test dataset. The evaluation consists of two scenarios: direct application of the object detection model on the blurred dataset and utilization of the object detection model alongside the Deblur model. Significant improvements in performance metrics of the YOLOv8 object detection model are observed when combined with Deblur model. These enhancements include a 7\% increase in recall, 7.7\% boost in accuracy, 5.4\% rise in mAP50, and a remarkable 11.6\% improvement in mAP50-95. Real-time validation confirms the superior performance of our proposed technique compared to the YOLOv8 baseline model. To validate this, a drone equipped with a portable camera captures FOD on the runway, showcasing improved object recognition capabilities across varying levels of image clarity. The broad use of this integrated system goes beyond runway FOD detection. This method may be used to any application that needs accurate detection of objects in low-visibility circumstances, such marine navigation, autonomous driving in bad weather, and surveillance in foggy conditions. Enhancing picture clarity prior to implementing detecting algorithms greatly increase reliability and accuracy.