FER to FFR: a deep-learning-based approach for robust fatigue detection

Automatic detection of fatigue from the face provides non-intrusive passive identification of fatigue. The traditional approach of fatigue detection has focused on detecting yawning and eyelid closure. However, fatigue is manifested in the face through various minute facial features. In this paper, we propose a fatigue detection model, which can learn facial expression features through a deep learning-based facial expression recognition model and provide the same to the fatigue recognition model. Experiments indicate that the proposed approach achieves a qualitative improvement of facial features used for fatigue detection and improves the accuracy quantitatively on the custom Indian fatigue data set. The approach also allows mitigation of limitations of fatigue data sets of significantly fewer subjects and allows for training fatigue models suitable for unconstrained real-world settings.