Détection de la somnolence au volant par ECG obtenu par des capteurs textiles capacitifs non intrusifs

Abstract

Road accidents represent a major issue, with a social cost of 1,000 dollars per capita in Canada in 2020, equivalent to 2% of the national GDP. Although technological and legislative advancements significantly reduced fatalities between 1973 and 2013, a concerning upward trend in fatal accidents has been observed since 2017. Among the identified causes, fatigue stands out as a factor in 24% of fatal accidents in Quebec, remaining one of the most challenging factors to reliably detect. In light of this challenge, physiological data, particularly electrocardiograms (ECG), show promise for drowsiness detection. However, their application in real-world conditions is hindered by technical limitations, such as the quality of signals captured capacitively or the invasiveness of traditional devices. This research explores the feasibility of a non-invasive drowsiness detection system based on capacitive ECG integrated into the driver’s seat. To achieve this goal, a capacitive textile system with three pairs of electrodes was developed, optimized, and evaluated with 25 participants. Initial results highlight a median performance of 83%(F1 score) for ECG peak detection. Additionally, a multimodal data collection over 75 hours of simulated driving allowed for the study of differences between wakefulness and drowsiness states. Classification performances reveal that capacitive ECG achieved an F1 score of 69.7 %, compared to 86.0 % for medical ECG. However, integrating physiological and vehicular data improved these results, with scores reaching 78.2 % for capacitive ECG and 88.6 % for medical ECG. These findings underscore the potential of capacitive ECG for drowsiness detection while highlighting challenges related to industrial integration. The best performances were obtained with small electrodes positioned in the middle of the back, and the multi-sensor system did not appear redundant : sensors had different optimal signal periods. This suggests that a system combining a grid of small electrodes and sensor fusion could enhance the robustness of this technology. This approach opens promising avenues for the development of non-invasive embedded safety systems, complementing existing solutions and improving the prevention of fatigue-related accidents.

Date	14 May 2025
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Ghyslain Gagnon (Supervisor) & Fabrice Vaussenat (Co-supervisor)