Residual Graph Network for Assessing Cognitive Fatigue from Gait Cycle Analysis

In this paper, we address the challenge of detecting cognitive fatigue, a condition marked by diminished cognitive capabilities during prolonged mental exertion, through a novel approach using the Residual Graph Convolutional Network (ResGCN) from the gait of individuals. Recognizing the limitations of traditional fatigue detection methods that rely on physiological signals like EEG, our research proposes a more accessible and unobtrusive technique by analyzing gait videos of individuals. Our model utilizes graph neural networks to encode the spatial positions of joint keypoints extracted from these videos via OpenPose, hypothesizing that cognitive fatigue manifests in altered body postures during walking. This approach is innovative in its use of non-contact data for fatigue detection, offering practicality and accessibility advantages over sensor-based methods. To train and test the proposed model, we use a custom gait dataset built from individuals before and after inducing cognitive fatigue. The proposed model achieves an accuracy of 85.7%, outperforming previous methods of detecting cognitive fatigue from gait cycle data.