Identification of Drunk People using Thermography and Machine Learning

Abstract

Ever since medical imaging and medical data analysis techniques have been discovered and put into practice, it has been one of the rapidly developing applications in the biomedical industry [1]. With various complex Artificial Intelligence (AI), machine learning, and deep learning models taking over, it is now possible to reconstruct or analyze any kind of complex conditions by training the model with a very dynamic and realistic dataset. Infrared Thermal Imaging (IRT) is one of the widely used types of medical imaging techniques that is used in various medical domains such as diagnosis of breast cancer, minute tumors, diabetes neuropathy, and other peripheral vascular disorders, etc [3]. By using the recent technologies, now it is possible to reproduce thermographic images with a higher quality which can be used to further innovate in real-time applications. Infrared thermal (IRT) imaging is a methodology that allows non-invasive and non-ionizing monitoring of skin surface temperature distribution, providing underlining physiological information on peripheral blood flow, autonomic nervous system, vasoconstriction/vasodilatation, inflammation, transpiration, or other processes that can contribute to skin temperature. It is a non-contact method that can be used to identify the superficial temperature of any object or surface[5]. This paper attempts to present a novel and safe, non-contact method to classify drunk and sober people, unlike the conventional techniques followed. Using a Breathalyzer during this COVID-19 situation is not a coherent approach that should be followed and can lead to high risks of infection as well. Other existing technologies which are present are only using anti-drunk driving systems which use the electrical signal impulses which pass from one’s heart to the brain [4]. Instead, by using the methods of thermal image processing, we can detect the drunk by analyzing the facial features (Figure 1). Thermal cameras have the capability to identify various features that indicates alcohol consumption. They are efficient enough to differentiate a temperature difference that is as small as 0.12 degrees Celcius even on a minute surface[6]. Studies have shown that the thermal behavior of the forehead and around the eyes widely increases when one is intoxicated with alcohol. This is because alcohol causes motor disturbances and also increases eye temperature [1].  To begin with, we gather an image dataset of people which includes the faces of both sober and drunk people who consumed alcohol in a wide range of quantities. The images are then appropriately augmented focussing mainly on the forehead and the eyes of the faces and further Face Detection process is been performed. When we train feature extraction using our proposed machine learning model, we will be able to identify a significant difference in the temperature while comparing a sober and drunk person. For future works, the same prototype can be infused into a real-time Thermal gun, which is similar to the Infrared Thermometers which are widely used in current practice since the pandemic situation. The device can be of great use for the Traffic Police to catch drunk drivers and can simultaneously generate the person’s Identity to verify their criminal history as we are going to integrate both Face recognition and Image Classification techniques here (Figure 2).