.png){kind=icon}
.png){kind=icon}
Return
Rethinking driver fatigue detection as anomaly identification: A hypergraph-transformer approach
.png){kind=icon}
PRE
AI
J
H
B
H
X
J
Y
DOI：10.1016/j.ipm.2025.104553.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Reframe driver fatigue detection as an anomaly detection problem, leveraging the natural observation that normal driving patterns are abundant while fatigue represents rare deviations from baseline behavior. • Introduce CBW dataset with 600+ h of synchronized vehicle CAN-bus and smartwatch data, the first multisource dataset designed specifically for anomaly-based fatigue detection using daily accessible sensors. • Develop Hypergraph-Transformer Driving Fatigue Detection (HG-TransDFD) framework with hierarchical hypergraph learning to capture complex cross-source relationships that traditional graph structures cannot effectively model. • Achieve superior performance with F1 scores of 0.71, 0.73, 0.75 on proposed CBW dataset, outperforming baseline methods by 5.9-6.3 % through effective multi-source data integration, while requiring only 10 % of labeled data to match traditional methods trained on 50 % labeled data, representing a 5x improvement in labeling requirements.
Journal
I
IF:
6.9
Papers:
5.2K
Citations:
1.4W
Organization
Cited Papers