DETECT DUI-IN CAR DRINK DRIVING DETECTION AND BACS

Abstract

As a significant contributor to road accidents and fatalities, the issue of drink driving warrants considerable research efforts. However, existing systems designed to detect or prevent drink driving often necessitate specialized hardware or significant user involvement, rendering them impractical for continuous monitoring in real-world driving scenarios. Addressing this challenge, we introduce DetectDUI, a non-invasive, contactless, and real-time system that offers a highly accurate approach to monitoring drink driving. Utilizing vital signs such as heart rate and respiration rate extracted from in-car WiFi signals, combined with the driver's psychomotor coordination inferred from steering wheel operations, DetectDUI presents a comprehensive framework.
The framework comprises a series of signal processing algorithms tailored to extract clean and informative vital signs and psychomotor coordination data. These data streams are then integrated using a self-attention convolutional neural network, known as C-Attention. In controlled laboratory experiments involving 15 participants, DetectDUI demonstrates remarkable efficacy, achieving a drink driving detection accuracy of 96.6% and accurately predicting blood alcohol concentration (BAC) levels with an average mean error ranging from 2 to 5mg/dl. These promising outcomes underscore the potential of DetectDUI as an effective tool for drink driving detection in real-world settings.