TY - GEN
T1 - Detection of prolonged stress in smart office
AU - Vildjiounaite, Elena
AU - Huotari, Ville
AU - Kallio, Johanna
AU - Kyllönen, Vesa
AU - Mäkelä, Satu Marja
AU - Gimel’farb, Georgy
N1 - Funding Information:
This work was supported by the Finnish Funding Agency for Technology and Innovation (Tekes) and VTT Technical Research Centre of Finland (ITEA 3 15008 ESTABLISH project).
Funding Information:
This research has been conducted as a part of the ITEA 3 15008 ESTAB-LISH project. We thank test subjects for their efforts to provide stress labels during a long study.
Publisher Copyright:
© Springer Nature Switzerland AG 2019.
Copyright:
Copyright 2019 Elsevier B.V., All rights reserved.
PY - 2018
Y1 - 2018
N2 - Detection of long-lasting stress in its early stage may allow prevention of irreversible health damages, but most of today’s stress detectors are not sufficiently convenient for long-term use. Some detectors rely on wearable physiological devices, known to have high abandonment rate, whereas other systems collect only behavioural data, but require each user to report occurrences of his/her stress during at least a month because behavioural patterns notably vary between individuals. Therefore, behaviour-based stress detection models should be learned separately for each person, and in existing systems such learning is typically supervised. In contrast, this work proposes a person-specific stress monitoring system, requiring no efforts from end users. This system acquires users’ motion trajectories via in-office depth cameras and employs a novel unsupervised method for analysis of these trajectories, based on discrete Hidden Markov Models. In 10-months-long real life study the proposed system correctly recognised the most stressful working periods of the monitored subjects and pointed out the most stress-prone persons.
AB - Detection of long-lasting stress in its early stage may allow prevention of irreversible health damages, but most of today’s stress detectors are not sufficiently convenient for long-term use. Some detectors rely on wearable physiological devices, known to have high abandonment rate, whereas other systems collect only behavioural data, but require each user to report occurrences of his/her stress during at least a month because behavioural patterns notably vary between individuals. Therefore, behaviour-based stress detection models should be learned separately for each person, and in existing systems such learning is typically supervised. In contrast, this work proposes a person-specific stress monitoring system, requiring no efforts from end users. This system acquires users’ motion trajectories via in-office depth cameras and employs a novel unsupervised method for analysis of these trajectories, based on discrete Hidden Markov Models. In 10-months-long real life study the proposed system correctly recognised the most stressful working periods of the monitored subjects and pointed out the most stress-prone persons.
KW - Ambient intelligence
KW - Hidden Markov Models (HMM)
KW - Stress detection
KW - Unsupervised learning
UR - http://www.scopus.com/inward/record.url?scp=85057106645&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-01177-2_90
DO - 10.1007/978-3-030-01177-2_90
M3 - Conference article in proceedings
AN - SCOPUS:85057106645
SN - 978-303-00117-6-5
T3 - Advances in Intelligent Systems and Computing
SP - 1253
EP - 1261
BT - Intelligent Computing
A2 - Kapoor, Supriya
A2 - Bhatia, Rahul
A2 - Arai, Kohei
PB - Springer
T2 - Computing Conference 2018, SAI 2018
Y2 - 10 July 2018 through 12 July 2018
ER -