Runway safety in aviation remains critical in maintaining passenger security and flight efficiency due to persistent risks such as incursions, foreign object debris (FOD), and unauthorized intrusions. Traditional safety monitoring systems depend on human monitoring and are often restricted to real-time responsiveness and adaptability to airport environments due to low visibility and high-traffic conditions. Thus, the research introduces an intelligent Network System for Runway Safety (INSy-RS), a deep learning-based framework for real-time detection, tracking, and safety analysis on active runways. The model applies You Only Look Once (YOLOv7) for efficient object detection to identify aircraft, ground vehicles, debris, and unauthorized personnel on runways. Deep Simple Online and Realtime Tracking (DeepSORT) is employed and integrated with spatio-temporal analysis to predict movement patterns and anomalous behaviors for multi-object tracking. The research model employs TensorRT optimization to achieve accelerated inference speeds on GPU hardware and guarantees real-time processing capabilities. The INSy-RS research results demonstrated that real-time runway surveillance improves detection accuracy across input resolutions by 15.41% compared to existing state-of-the-art models. With the 25.33% increase in inference speed, conflict precision increases by 56.94% and mitigates incursion alert time by 39.77%, allowing faster runway violation response. With 5.42% higher mAP@0.5, INSy-RS improves runway object and entity detection and tracking. The proposed research model has a practical application in airport control towers and autonomous surveillance drones, thus improving situational awareness and response efficiency to potential runway threats.
Zhang et al. (Tue,) studied this question.