TY - JOUR
T1 - Regression-based continuous driving fatigue estimation
T2 - Toward practical implementation
AU - Bose, Rohit
AU - Wang, Hongtao
AU - Dragomir, Andrei
AU - Thakor, Nitish V.
AU - Bezerianos, Anastasios
AU - Li, Junhua
N1 - Funding Information:
Manuscript received October 16, 2018; revised April 4, 2019; accepted July 1, 2019. Date of publication July 19, 2019; date of current version June 10, 2020. This work was supported in part by the National Natural Science Foundation of China under Grant 61806149, in part by the Natural Science Foundation of Guangdong Province under Grant 2018A030313882, in part by the Projects for International Scientific and Technological Cooperation under Grant 2018A05056084, in part by the Ministry of Education of Singapore under Grant MOE2014-T2-1-115, in part by the NSU Startup under Grant R-719-000-200-133, and in part by the Science Foundation for Young Teachers of Wuyi University under Grant 2018td01. (Corresponding author: Junhua Li.) R. Bose and N. V. Thakor are with the Singapore Institute for Neurotechnology, National University of Singapore, Singapore.
Publisher Copyright:
© 2016 IEEE.
PY - 2020/6
Y1 - 2020/6
N2 - Mental fatigue in drivers is one of the leading causes that give rise to traffic accidents. Electroencephalography (EEG)-based driving fatigue studies showed promising performance in fatigue monitoring. However, complex methodologies are not suitable for practical implementation. In our simulation-based setup that retained the constraints of real driving, we took a step closer to fatigue estimation in a practical scenario. We adopted a preprocessing pipeline with low computational complexity, which can be easily and practically implemented in real time. Moreover, regression-based continuous fatigue estimation was achieved using power spectral features in conjunction with time as the fatigue label. We sought to compare three regression models and three time windows to demonstrate their effects on the performance of fatigue estimation. Dynamic time warping was proposed as a new measure for evaluating the performance of fatigue estimation. The results derived from the validation of the proposed framework on 19 subjects showed that our proposed framework was promising toward practical implementation. Fatigue estimation by the support vector regression with radial basis function kernel and 5-s window length achieved the best performance. We also provided a comprehensive analysis on the spatial distribution of channels and frequency bands mostly contributing to fatigue estimation, which can inform the feature and channel reduction for real-time fatigue monitoring in practical driving. After reducing the number of electrodes by 75%, the proposed framework retained comparable performance in fatigue estimation. This paper demonstrates the feasibility and adaptability of our proposed framework in practical implementation of mental fatigue estimation.
AB - Mental fatigue in drivers is one of the leading causes that give rise to traffic accidents. Electroencephalography (EEG)-based driving fatigue studies showed promising performance in fatigue monitoring. However, complex methodologies are not suitable for practical implementation. In our simulation-based setup that retained the constraints of real driving, we took a step closer to fatigue estimation in a practical scenario. We adopted a preprocessing pipeline with low computational complexity, which can be easily and practically implemented in real time. Moreover, regression-based continuous fatigue estimation was achieved using power spectral features in conjunction with time as the fatigue label. We sought to compare three regression models and three time windows to demonstrate their effects on the performance of fatigue estimation. Dynamic time warping was proposed as a new measure for evaluating the performance of fatigue estimation. The results derived from the validation of the proposed framework on 19 subjects showed that our proposed framework was promising toward practical implementation. Fatigue estimation by the support vector regression with radial basis function kernel and 5-s window length achieved the best performance. We also provided a comprehensive analysis on the spatial distribution of channels and frequency bands mostly contributing to fatigue estimation, which can inform the feature and channel reduction for real-time fatigue monitoring in practical driving. After reducing the number of electrodes by 75%, the proposed framework retained comparable performance in fatigue estimation. This paper demonstrates the feasibility and adaptability of our proposed framework in practical implementation of mental fatigue estimation.
KW - Driving fatigue estimation
KW - Dry electrode
KW - Dynamic time warping (dtw)
KW - Electroencephalography (eeg)
KW - Regression
KW - Wireless transmission
UR - http://www.scopus.com/inward/record.url?scp=85071036774&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85071036774&partnerID=8YFLogxK
U2 - 10.1109/TCDS.2019.2929858
DO - 10.1109/TCDS.2019.2929858
M3 - Article
AN - SCOPUS:85071036774
SN - 2379-8920
VL - 12
SP - 323
EP - 331
JO - IEEE Transactions on Cognitive and Developmental Systems
JF - IEEE Transactions on Cognitive and Developmental Systems
IS - 2
M1 - 8766845
ER -