Understanding Influences of Driving Fatigue on Driver Fingerprinting Identification Through Deep Learning

• Type: Article in International Scientific Journal
• Date: 2024

Authors

Yifan Sun (Author) Other; Chaozhong Wu (Author) Other; Hui Zhang (Author) Other; Sara Ferreira (Author) FEUP; José Pedro Tavares (Author) FEUP; Naikan Ding (Author) Other

Journal

Title: IEEE Transactions on Vehicular Technology

Vol. 73

Pages: 1829-1844

ISSN: 0018-9545

Publisher: IEEE

Indexing

ISI Web of Knowledge - 0 Citations

ISI Web of Science

Scopus

Google Scholar

Clarivate Analytics

Scientific classification

CORDIS: Technological sciences > Engineering > Civil engineering

FOS: Engineering and technology > Civil engineering

Other information

Authenticus ID: P-010-CVP

DOI: 10.1109/tvt.2023.3320679

Abstract (EN): Driver Fingerprinting (DF), reflecting unique driving characteristics, has received extensive attention for its prominent applications in various domains such as driver identification in automobile sharing paradigms, driver-based insurance, and vehicle anti-theft. Previous research has primarily concentrated on utilizing data from sober states to establish DF models while rarely considering driving fatigue, a common and risky driving state. This study proposes a scheme to analyze the effects of fatigue on DF identification and emphasizes the necessity of incorporating fatigue in DF models. Firstly, we conducted simulated driving experiments to collect the drivers' behavior data, facial videos, and the Karolinska Sleepiness Scale (KSS). Secondly, we calculated DF indicators using a double-layer sliding time window and used the Kruskal-Wallis test to extract the DF features. Finally, we divided the full sample set into three subsets: a sober sample subset, a fatigue sample subset, and a double state sample subset containing both sober and fatigue data. We used each sample subset to train DF models based on principal component analysis and long short-term memory. We then utilized all three sample subsets to evaluate the DF models trained by each subset. Combined with the correlation between the indicators and KSS, we analyzed the influence of fatigue on DF identification. The results indicated that the identification accuracies of DF models built solely on sober data significantly decreased when applied to fatigue data. The average accuracy of the 7-driver groups showed the most substantial reduction of 54.59%. Conversely, DF models considering fatigue demonstrated a notable increase in the identification accuracy of drivers, and the maximum increase of 25.47% was from the 7-driver groups. Additionally, we discussed the effects of time windows and the number of drivers on DF identification accuracy and an application framework for driver identification systems based on DF integrating driving fatigue detection.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 16