A Deep-Learning-Enabled Monitoring System for Recognizing G-induced Loss of Consciousness

Loss of consciousness induced by persistent positive acceleration is a significant hidden danger threatening the safety of maneuvering flight, necessitating timely and accurate monitoring and warnings. However, traditional monitoring systems for loss of consciousness primarily rely on electrophysiological signals, which are affected by vibration, electromagnetic interference, and other factors during actual flight. Consequently, these measurements are prone to external interference and lack robustness. To address these issues, we propose an automatic loss-of-consciousness monitoring system (ALOCMS) based on computational vision. This system utilizes the time-series features of the pilot’s head posture to identify the pilot’s state of consciousness effectively. To enhance the model’s generalization ability, we built a head pose dataset specific to flight scenarios and merged it with a public dataset for training the network model. Additionally, we developed a user-friendly human-computer interface to display monitoring information intuitively. Experimental results demonstrate that this system can accurately recognize and alert loss of consciousness in real time.