G5 Doctoral dissertation (article)
Automatic detection of driver’s abnormal state based on physiological signals (2024)
Automaattinen kuljettajan poikkeavan tilan havaitseminen fysiologisten signaalien perusteella

Zuo, X. (2024). Automatic detection of driver’s abnormal state based on physiological signals [Doctoral dissertation]. University of Jyväskylä. JYU dissertations, 765. https://urn.fi/URN:ISBN:978-952-86-0099-2

JYU authors or editors

Publication details

All authors or editorsZuo, Xin


Journal or seriesJYU dissertations


Publication year2024

Number in series765

Number of pages in the book1 verkkoaineisto (72 sivua, 36 sivua useina numerointijaksoina, 4 numeroimatonta sivua)

PublisherUniversity of Jyväskylä

Publication countryFinland

Publication languageEnglish

Persistent website addresshttps://urn.fi/URN:ISBN:978-952-86-0099-2

Publication open accessOpenly available

Publication channel open accessOpen Access channel


Abnormal driver state affecting environment perception, decision-making, and actions is one of the main traffic accidents causes. Physiological signals, reflecting drivers’ actual internal state, have been used to detect abnormal status. Time and frequency domain features are commonly adopted to study driver state, while they are sensitive to residual noise and neglect signals’ complexity. Besides, high temporal resolution is necessary to detect and analyze the subtle changes in driver status timely at a certain time but increases the sample rate and may decrease the computational efficiency. In long-term driving operations, the long-term temporal dependency is also vital for the automatic detection of the diver’s abnormal state. However, difficulties exist in optimizing sample rate considering the time complexity of physiological signals and detecting abnormal driver status automatically with respect to long-term context information.
This dissertation focuses on the above challenges and proposes to detect the driver’s abnormal state with multiscale entropy of physiological signals and bidirectional long short-term memory (BiLSTM) network. Article I explores the complexity of electroencephalography (EEG) with multiscale entropy on relative time scales (MSE) and the information compensation manner among features in detecting distraction. Article II investigates the fluctuation patterns of MSE and considers the long-term dependency on features with BiLSTM for distraction detection. Article III studies the appropriate sample rate by calculating multiscale entropy on absolute time scales (MSaE) and explores the distraction information in multiple physiological signals to detect distraction. Articles II and III also analyze the behavioral signals to validate the changes in driving performance due to distraction. In Article IV, a cross-subject emotion recognition framework based on fused entropy features and BiLSTM is proposed to integrate the merits of different features and learn the contextual information in EEG.
In summary, this dissertation investigates the fluctuation patterns of physiological signals with multiscale entropy of the optimized sample rate under different mental statuses to detect abnormal states with BiLSTM. The proposed framework indicates the potential of understanding and detecting a driver’s abnormal state with multiple signals.

Keywordsdrivers (occupations)ability to drivetraffic behaviourmemory (cognition)interferenceschangephysiological effectssignalssignal analysisentropydoctoral dissertations

Free keywordsmultiscale entropy; long short-term memory; multi-modality analysis

Contributing organizations

Ministry reportingYes

Reporting Year2024

Last updated on 2024-13-05 at 18:45