Recognition method of fatigue state and unsafe behavior of crane drivers based on computer vision

To improve the safety and efficiency of tower crane operations, this study proposes a comprehensive identification method for fatigue and unsafe behaviors, which can timely detect and identify the possible fatigue states and unsafe behaviors of drivers. This method uses a camera to capture real-time video streams, and preprocesses and analyzes the videos to extract key information for subsequent fatigue and unsafe behavior recognition. A recognition method based on eye and mouth status is adopted for fatigue state, analyzing indicators such as eye opening and closing status, blink frequency, and yawning frequency. Unsafe behavior recognition adopts computer vision and deep learning methods to identify in real-time the dangerous operations that drivers may perform, such as using mobile phones, smoking, etc. The results showed that after model optimization, the YOLOv5-ECA model exhibited significant performance improvement in fatigue state and unsafe behavior recognition, with accuracy and recall rates exceeding 90%. Verify the high accuracy of the model in identifying different categories through visual analysis results.