Application of Deep Learning for Ergonomic Data Augmentation and Human State Recognition

Abstract

In ergonomic experiments, a small number of participants is often a problem because a sufficient amount of data is not obtained. In recent years, human state recognition is wide-spread, and estimating the human state from biological information acquired from a wearable device, is useful for improving living behavior. While it is necessary to collect a sufficient amount of data in order to perform state estimation with a certain degree of accuracy, collecting the amount of data requires a considerable cost. This study attempted to expand physiological and psychological data using deep learning. Specifically, information on physiological indicators was added to ACGAN. From the verification using the actual experimental results, it was found that the accuracy of recognizing the human state was improved by using the augmented data compared to the case of learning with a small number of original data.