Toshiki Hayashida (Graduate School of Information Science and Electrical Engineering)’s paper has been accepted for Journal of Information Processing.
Congratulations!
Authors
Toshiki Hayashida, Shogo Fukushima, Yugo Nakamura, Hyuckjin Choi, Yutaka Arakawa
Affiliation
Graduate School of Information Science and Electrical Engineering,
Department of Information Science and Technology
Manuscript Title
Can a Chair Detect Nodding?: An Exploratory Study for Privacy-Aware Meeting Analysis
Abstract
In this paper, we explore a novel non-intrusive nodding recognition method for privacy-awaremeeting analysis. We focus on the subtle sway through the chair on which a participant is seated, employingan accelerometer mounted to the chair’s backrest to recognize nodding. By capturing subtle sway of the chairrather than using cameras or wearable devices, our method preserves participants’ privacy while enablingnaturalistic nodding measurement in real meeting settings. In experimental evaluations conducted undercontrolled conditions, a machine learning model based on Random Forest(RF) achieved nodding recognitionaccuracy exceeding 98%. Nod count accuracy was also favorable, with a mean absolute error (MAE) of 5.40and a weighted absolute percentage error (WAPE) of 7.11%. In a subsequent evaluation in an actual meetingenvironment, recognition accuracy decreased due to body movements and posture changes; nevertheless, theresults suggest the feasibility of nodding recognition via chair sway sensing. Our results suggest a practical,privacy-aware route to unobtrusive meeting analysis via chair-sway sensing.
Journal name
Journal of Information Processing
Relevant SDGs
SDGs 3 (Good Health and Well-Being), SDGs 8(Decent Work and Economic Growth)
Comments
This paper presents a method for recognizing nodding, a nonverbal behavior commonly observed during meetings, without relying on cameras or wearable devices. The key idea is to use an accelerometer attached to the backrest of a chair to capture the subtle chair sway caused by a participant’s nodding, and to estimate both the presence and frequency of nods using machine learning techniques. Experimental results under controlled conditions demonstrated a recognition accuracy exceeding 98% using a single backrest-mounted sensor. In addition, the number of nods could be estimated with an error rate of approximately 7%. These findings suggest that it is possible to recognize and quantify participants’ nodding behavior from chair sway while preserving privacy and minimizing user burden.
