Controller fatigue state detection based on improved YOLOv5

To improve the real-time monitoring ability of air traffic controllers, here this paper introduces a novel approach for detecting fatigue through facial recognition technology. The YOLOv5s algorithm is critically improved by integrating CBAM attention mechanism, BiFPN structure and GSConv convolution, significantly enhancing the accuracy and efficiency of face detection. On the basis of successful face detection, ERT algorithm is further applied to accurately locate facial feature points, and by calculating eye aspect ratio (EAR) and percent eye closure (PERCLOS), combined with P80 standard, the fatigue state is accurately judged. The results indicate that the improved YOLOv5 model has achieved a $\mathbf{1 . 2 \%}$ improvement in accuracy, a $0.1 \%$ increase in mean Average Precision (mAP), and a $0.5 \%$ rise in recall rate over the initial algorithmic framework. In summary, the detection precision of the refined model has ascended to $\mathbf{9 7 . 8} \%$. The improved YOLOv5 model can effectively detect fatigue state.