We recently published a paper at the 2023 CHI Conference on Human Factors in Computing Systems, presenting an in-vehicle machine learning system designed to predict critical blood alcohol concentration (BAC) levels in real-time. This system leverages driver monitoring cameras—now mandated in many countries—to assess a driver’s BAC.
Our evaluation involved a simulator study with 30 participants, demonstrating that the system reliably detected any alcohol influence while driving with an area under the ROC-curve (AUROC) of 0.88. It also identified drivers exceeding the WHO-recommended limit of 0.05 g/dL BAC with an AUROC of 0.79. Model inspection revealed that the system relies on pathophysiological effects associated with alcohol consumption.
This research represents the first rigorous evaluation of driver monitoring cameras for detecting drunk driving, highlighting their potential to prevent alcohol-related harm.
Technical Details
The system combines visual and behavioral cues captured by in-vehicle cameras to predict BAC levels. Key features include:
- Facial landmarks
- Eye movements
- Driving behavior patterns
The machine learning model was trained on data from the simulator study, ensuring robust performance in real-world scenarios. The AUROC values indicate high accuracy in detecting critical BAC levels, making this system a promising tool for enhancing road safety.
📖 Learn more in the original research paper →.
Citation
Koch, K., Maritsch, M., van Weenen, E., Feuerriegel, S., Pfäffli, M., Fleisch, E., Weinmann, W., Wortmann, F. 2023. Leveraging driver vehicle and environment interaction: Machine learning using driver monitoring cameras to detect drunk driving. Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, 1–32. https://doi.org/10.1145/3544548.3580975