Digital filtering of the pupil diameter signal for affective assessment of a computer user

Abstract

There is an increasing interest in investigating means to detect the level of stress of a computer user for applications such as computer-based tutoring. This study presents a new approach to recognize the affective state (relaxed vs. stressed) of a computer user through filtering of the pupil diameter (PD) signal. Wavelet denoising and Kalman filtering are applied for the removal of abrupt changes in the PD signal, which are not likely to be the actual pupillary reactions. Wavelet denoising comprises three steps: Wavelet decomposition, threshold setting and signal reconstruction. Kalman filtering includes prediction and correction applied as two recursive blocks. This study summarizes the preliminary results obtained for noise removal and imposes a threshold on the filtered signal to identify the "stress" state of the computer users. Direct classification applied on the filtered PD signals achieved 75 % or better accuracy for stress detection in 3/4 of the subjects tested. These encouraging results indicate that the combination of wavelet denoising and Kalman filtering seems to be an appropriate processing approach to emphasize the PD signal changes due to affective variations in the subject. The results also show that the PD signal is an important physiological signal that can be monitored non-invasively for the affective assessment of a computer user. © 2013 Springer Science+Business Media.