Deep Learning Technology for Drunks Detection with Infrared Camera

Abstract

This project manipulates for the purpose to solve the problem of alcohol measurement delays and reduces the issue of contamination that arises from the breath analyzer test. This project focuses on designing a novel infrared (IR) camera-based alcohol detection system with deep learning technology. This system consists of 2 parts. The first part is an infrared camera (FLIR) used for collecting both IR and normal images then the next part is an image processing system for alcohol detection based on deep learning technology operating on an iPhone operating system (iOS) mobile phone. Our handheld IR based detection system achieves an accuracy of 85.10% (135 population) accuracy with 4 levels of classification (sober, 1 glass, 2 glasses, or 3 glasses) and 74.07% with binary identification (Sober or Drunk). Each glass contains 200 ml of beer (5% vol).