For several decades, numerous reports have highlighted the role of drowsy drivers in fatal accidents. Institutions are therefore seeking new ways to limit the impact of drowsiness on road safety. This interest in reducing driving-related fatalities has prompted researchers to develop new, non-invasive methods for monitoring drivers to detect drowsiness.
The main objective of this thesis is to propose a new non-invasive cardio-respiratory acquisition system based on textile inductive electrodes to detect drowsiness and improve the accuracy of this type of system. This thesis focuses on the respiratory signal. To achieve this, an innovative method using four electrodes distributed along the seat belt and fusion through a Kalman filter, which estimates the future state of a system by reducing data uncertainty, was developed. To evaluate the system, different electrode positions and sizes were tested, and their performance was measured in a simulated driving situation.
The results demonstrate that electrode position significantly impacts accuracy, with electrodes positioned on the seat belt achieving better performance than those on the back. Larger electrodes were found to maximize accuracy, and generally, for optimal signal quality, the relative movement of electrodes with respect to the subject should be minimized. Over the 50 sessions of 1.5 hours, using the Kalman filter increased the rate of recordings with a mean absolute respiratory rate estimation error of less than 2 breaths per minute from 28% to 44%. The drowsiness classification accuracy achieved with this textile system is comparable to that obtained with reference data.
In conclusion, contactless textile inductive electrodes are a promising option for respiratory monitoring and drowsiness detection in driving conditions. However, improvements are still needed, such as a matrix of textile electrodes directly embroidered onto the seat belt to limit their movement. Future studies should focus on improving quality indices to detect poor-quality signals and enhance drowsiness detection accuracy by minimizing the impact of noise. Acquiring new data would improve drowsiness detection performance by enhancing model generalization.
| Date | 12 Dec 2024 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Ghyslain Gagnon (Supervisor) & Fabrice Vaussenat (Co-supervisor) |
|---|
Bellemin, V. (Author),
Gagnon (Supervisor) & Vaussenat (Co-supervisor),
12 Dec 2024Student thesis: Master's thesis › Master in Engineering: Electrical Engineering