Alcohol Consumption Detection from Periocular NIR Images Using Capsule Network

TL;DR

This paper introduces a novel Fused Capsule Network for detecting alcohol consumption from periocular NIR images, achieving high accuracy and potentially aiding in safety assessments.

Contribution

It presents a new Fused Capsule Network architecture specifically designed for classifying iris NIR images affected by alcohol consumption, with improved efficiency.

Findings

Achieved 92.3% accuracy in detecting alcohol use from iris NIR images.

F-CapsNet uses half the parameters of standard Capsule Networks.

Demonstrates potential for automatic alcohol detection systems.

Abstract

This research proposes a method to detect alcohol consumption from Near-Infra-Red (NIR) periocular eye images. The study focuses on determining the effect of external factors such as alcohol on the Central Nervous System (CNS). The goal is to analyse how this impacts on iris and pupil movements and if it is possible to capture these changes with a standard iris NIR camera. This paper proposes a novel Fused Capsule Network (F-CapsNet) to classify iris NIR images taken under alcohol consumption subjects. The results show the F-CapsNet algorithm can detect alcohol consumption in iris NIR images with an accuracy of 92.3% using half of the parameters as the standard Capsule Network algorithm. This work is a step forward in developing an automatic system to estimate "Fitness for Duty" and prevent accidents due to alcohol consumption.