Project
| # | Title | Team Members | TA | Documents | Sponsor |
|---|---|---|---|---|---|
| 28 | Real-time EEG Drowsiness Detection Device |
Nikhil Talwalkar Senturran Elangovan |
Zhuoer Zhang | proposal1.pdf |
|
| **Real-time EEG Drowsiness Detection Device** Team members: - Nikhil Talwalkar (nikhilt4) - Senturran Elangovan (se10) **Problem** Many people unintentionally doze off while studying, working, or in situations that demand constant focus—such as driving or monitoring critical systems. Current consumer sleep trackers, such as smartwatches, are primarily designed to analyze and record sleep patterns after the fact. They cannot provide real-time interventions to prevent drowsiness-related lapses. In high-risk scenarios like long-distance or nighttime driving, even a few seconds of microsleep can result in serious accidents. Therefore, there is a need for a portable, proactive system that can detect drowsiness in real time and alert users before loss of focus occurs. **Solution** Our project proposes an implementation of a real-time drowsiness detection device. The system uses a lightweight EEG headband to continuously monitor the user's brain activity. By analyzing frequency changes in EEG signals associated with early stages of drowsiness, the device can detect when the user is at risk of falling asleep. When drowsiness is detected, the system triggers an audible or tactile alarm to immediately alert the user, helping prevent microsleep-related accidents or lapses in attention. Compared to computer vision–based systems, which rely on slower external cues such as eyelid closure, yawning, or head movement, and which often perform poorly in nighttime conditions, our device provides earlier and more reliable detection by directly monitoring EEG signals. To ensure usability and a practical aesthetic, the electrodes will be put into a cap-style wearable that requires so special alignment or positioning by the user. The device will be powered by a lithium polymer battery with a projected life of 8 to 10 hours. In terms of performance, most false positives and false negatives arise from interference in EEG signals, such as when eyes are closed during meditation or half-open. Since these states are not relevant to driving scenarios, they will still trigger an alert. We expect a 1–2% false positive rate during normal focus and a less than 10% false negative rate when the user is drowsy. **Subsystems:** **Subsystem 1: EEG Headband Hardware** - Lightweight, dry electrodes attached to Fp1, Fp2, and Fpz regions of the head, wired neatly into a cap-style wearable to capture brain activity. - Ideally dry, reusable, Ag electrodes, restricted to the budget. If allowed, higher end electrodes can be integrated for future modifications. **Subsystem 2: Signal Processing Unit** - Analog noise filtering using Butterworth filter, aiming for a bandpass between 0.5 to 30 Hz. - Includes a CMRR operational amplifier to amplify the signal from 10^{-3} range to 1. - Analog to digital signal converter to allow signal to be filtered digitally for flexibility and data collection. f_{s} around 250Hz for good signal. **Subsystem 3: Detection Algorithm** - Software running locally to identify characteristic frequency changes in EEG that correspond to drowsiness - Using open libraries such as MNE, YASA and others for the EEG signal processing - ML (RFA, Naive Bayes) algorithms to determine if user is at the brink of stage 1 sleep **Subsystem 4: Alert Mechanism** - Audible (buzzer) or tactile (vibration motor) alerts to immediately notify the user when drowsiness is detected. **Subsystem 5: Power System** - Lithium-polymer battery providing 8–12 hours of continuous operation for portability and reliability - Power electronic circuit to ensure battery doesn't overcharge or over-discharge, and maintain limited current draw, and if possible (time constraints) temperature monitoring. - Exploring 'AAA' battery alternatives if integratabtle and doesn't make the device look too chunky. **Criterion of Success:** EEG acquisition – the EEG captures reliable and accurate brain signals. Can be checked by blinking, which will induce a relatively significant voltage spike in the EEG signal. Real-time sleep detection – the control system can detect when the user feels has micro-sleeps or is drowsy. Feeding open-source data or sleep and drowsiness into the system, and check if there is any outputs. Prompt alerting – the buzzer triggers the alerting noise at a timely manner, with acceptable detection-to-alert latency. Measuring the time delay from the input and the output signal, to ensure the latency is acceptable. Safety and comfort – the device is wearable for long hours and safe. Since user-based, allow randomly-selected volunteers to wear for a day and tell if there's any discomfort. Quantize it by using a numbering survey. TA can be included too, if they volunteer. **Resources:** https://github.com/SuperBruceJia/EEG-DL https://github.com/lcsig/Sleep-Stages-Classification-by-EEG-Signals https://www.sciencedirect.com/science/article/pii/S2090447922002064 Bohao Li, et al 2021, J. Phys.: Conf. Ser. 1907 012045 |
|||||