An Enhanced OpenPose-Based Methodology for Detecting Pilot Fatigue During Flight Operations

Abstract

Single-Pilot Operations (SPO) represent a focal point in the current development direction of commercial aircraft, where monitoring the workload of a single pilot is crucial for ensuring flight safety. The detection results must exhibit both accuracy and real-time performance. This study leverages the widely adopted OpenPose model, which is employed for posture detection due to its capability to effectively extract human skeletal keypoints. However, the conventional OpenPose model suffers from issues such as large data volume, high computational demands, and significant redundancy. This paper systematically reviews relevant improvement methods across three key aspects of the OpenPose model: image preprocessing, feature extraction, and pose estimation. Additionally, it summarizes a fatigue assessment method based on this model. Research findings indicate that the enhanced OpenPose model, compared to the traditional version, demonstrates greater adaptability to complex environments, faster response and data processing capabilities, and improved accuracy in results. By incorporating fatigue detection criteria, the model enables real-time alerting.