Leveraging driver vehicle and environment interaction: Machine learning using driver monitoring cameras - drunk driving detection preprint

Aug 2, 2026

Upload by Logic

Excessive alcohol consumption leads to disability and death, and digital interventions can reduce alcohol-related harm by enabling real-time behavior change during critical moments like driving. The work presents an in-vehicle machine learning system that predicts critical blood alcohol concentration levels using driver monitoring cameras mandated in many countries. An interventional simulator study with 30 participants evaluates detection accuracy. The approach identifies both any alcohol influence and BAC above WHO’s 0.05 g/dL threshold, with model inspection linking decisions to alcohol-related pathophysiological effects.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages32

Reading Time1 h 21 min

AccessFree to download

Abstract

Introduction

Background: alcohol harm and need for real-time predictions

Limitations of ignition interlock devices and cost challenges

Proposed approach using driver monitoring cameras

Contributions

Preprint-Leveraging driver vehicle and environment interaction: Machine learning using driver monitoring cameras to detect drunk driving arXiv :2301 .08978v1 [ cs .HC] 21 Jan 2023 Kevin Koch∗ kevin.koch@unisg.ch University of St. Gallen St. Gallen, Switzerland Stefan Feuerriegel [feuerriegel@lmu.de](feuerriegel@lmu.de)[ ](feuerriegel@lmu.de)LMU Munich Munich, Germany Martin Maritsch∗ mmaritsch@ethz.ch ETH Zürich Zürich, Switzerland Matthias Pfäffli [matthias.pfaeffli@irm.unibe.ch](matthias.pfaeffli@irm.unibe.ch)[ ](matthias.pfaeffli@irm.unibe.ch)University of Bern Bern, Switzerland Eva van Weenen [evanweenen@ethz.ch](evanweenen@ethz.ch)[ ](evanweenen@ethz.ch)ETH Zürich Zürich, Switzerland Elgar Fleisch efleisch@ethz.ch ETH Zürich and University of St. Gallen Zürich and St. Gallen, Switzerland Wolfgang Weinmann† [wolfgang.weinmann@irm.unibe.ch](wolfgang.weinmann@irm.unibe.ch)[ ](wolfgang.weinmann@irm.unibe.ch)University of Bern Bern, Switzerland ABSTRACT Excessive alcohol consumption causes disability and death. Digital interventions are promising means to promote behavioral change and thus prevent alcohol-related harm, especially in critical moments such as driving. This requires real-time information on a person’s blood alcohol concentration (BAC) . Here, we develop an in-vehicle machine learning system to predict critical BAC levels. Our system leverages driver monitoring cameras mandated in numerous countries worldwide. We evaluate our system with 􀀽 = 30 participants in an interventional simulator study. Our system reliably detects driving under any alcohol influence (area under the receiver operating characteristic curve [AUROC] 0. 88) and driving above the WHO recommended limit of 0. 05 g/dL BAC (AUROC 0. 79) . Model inspection reveals reliance on pathophysiological effects associated with alcohol consumption. To our knowledge, weare the first to rigorously evaluate the use of driver monitoring cameras for detecting drunk driving. Our results highlight the potential of driver monitoring cameras and enable next-generation drunk driver interaction preventing alcohol-related harm. CCS CONCEPTS • Human-centered computing → Empirical studies in HCI; Ubiquitous and mobile computing systems and tools; • Applied computing → Consumer health. KEYWORDS health; safety; driving; alcohol; eye movements; head movements; driver monitoring ∗Joint first author. †Joint last author.‡Corresponding author. Felix Wortmann†‡ [felix.wortmann@unisg.ch](felix.wortmann@unisg.ch)[ ](felix.wortmann@unisg.ch)[University of St. Gallen](University of St. Gallen) St. Gallen, Switzerland 1 INTRODUCTION Alcohol consumption is responsible for 5% of the global disease burden and is further the cause of 1 in 20 deaths worldwide [110] . To promote behavior change, digital interventions provide effective means to prevent harm in critical situations due to alcohol consumption and intoxication [3, 5, 64, 65] . In particular, digital interventions could promote behavior change by delivering real-time targeted feedback on alcohol consumption. However, to intervene early, real-time predictions of alcohol consumption are needed. Alcohol consumption increases, among others, the risk of traffic crashes, making drunk driving one of the leading causes of severe crashes on public roads. For example, in the US, around 30 people die each day in traffic crashes in which one of the parties is under the influence of alcohol, and, together, alcohol-related crashes amount to 30% of all traffic fatalities [67]. To prevent alcohol-related crashes, in-vehicle systems are needed to detect drunk driving and enable targeted interventions. Examples of such interventions are, e.g., warnings of impairment and forced stops ofthe vehicle. As of today, the only reliable measurement technology for identifying intoxicated driving are ignition interlock devices that analyze a driver’s breath alcohol. However, ignition interlock devices

FAQs

What problem does the paper address?

The paper targets detecting drunk driving by predicting critical blood alcohol concentration levels in real time while someone drives.

How does the proposed system estimate alcohol influence?

It uses driver monitoring cameras to extract gaze behavior and head movements, then predicts whether drivers exceed BAC thresholds.

How is the system evaluated?

The system is evaluated in an interventional simulator study with 30 participants, reporting AUROC values for detecting any alcohol influence and BAC above the WHO-recommended limit.

Aug 2, 2026

Upload by Logic