Identification Of Drinking Intoxication: Applying Artificial Intelligence For Improved Traffic Safety

Document Type

Article

Publication Date

1-1-2026

School

Computing Sciences and Computer Engineering

Abstract

Drunk driving detection systems are traditionally reactive, relying on tools such as breathalyzers or field sobriety tests deployed only after unsafe behavior is observed. To address this limitation, recent advances in computer vision and deep learning have enabled the development of proactive, non-intrusive systems capable of detecting intoxication based on facial features. This paper expands on our previous work by focusing on out-of-vehicle drunk driving detection, leveraging facial imagery captured from external sources such as roadside cameras or drones. We apply Machine Learning (ML) and Deep Learning (DL) models to a large-scale dataset of sober and intoxicated faces, introducing controlled salt-and-pepper noise at 20%, 40%, and 50% levels, along with disruption techniques such as flipping and brightness variations to simulate real-world surveillance conditions. We proposed two configuration modes, low-resource and high-resource models, to illustrate the applicability of our scheme to devices with different resource constraints. To enhance transparency, we integrate Explainable AI (XAI) tools—such as saliency maps—to identify key facial regions influencing model decisions. In addition to software-based evaluation, this work investigates the feasibility of real-time deployment through a hardware–software co-design implemented on a Field-Programmable Gate Array (FPGA) platform. Both low-resource and high-resource model configurations are analyzed with respect to architectural design and resource utilization, demonstrating the practicality of embedded inference in roadside and edge environments. By supporting efficient edge-level inference, the proposed system is well-suited for Internet-of-Things (IoT) deployments that rely on distributed roadside sensors and embedded processing platforms.

Publication Title

IEEE Internet of Things Journal

Link to Full Text

Share

COinS