Recognizing Confusion in Assembly Work based on a Hidden Markov Model of Gaze Transition

A method to detect the states of confusion during an assembly work for a user-adaptive work support system is proposed in this study. The eye-gaze data from an eye tracker are discretized into predefined areas of interest, and a hidden Markov model is used to classify the cognitive states associated with confusion and types of work steps involved. An offline experiment using data collected from 16 participants showed low task dependency (F1-scores of 0.638 and 0.639 for dependent and independent evaluations, respectively) and relatively high person dependency (F1-score of 0.621 for independent evaluation). The high discriminability between the two assembly steps and tendency of misclassification in confused states were confirmed.