Drunken4VN: A Thermal Infrared Imaging Dataset for Alcohol Intoxication Detection in Vietnamese under Diverse Environmental Conditions

Alcohol intoxication is an international issue of public health and safety, a major cause of road accidents, workplace dangers, and avoidable fatalities. While the subjects of tests like breath alcohol testing need cooperation, blood tests are cumbersome, yet there is growing demand for invasive-free, real-time sensors to detect intoxication, and existing research points to facial temperature pattern, captured through thermal imaging, as an effective behavioral physiology and marker of intoxication. Nevertheless, we have no public-access datasets of alcohol detection by thermal imaging, in particular for Southeast Asian populations whose environmental and physiological properties may differ from other populations. In an attempt to fill this gap, we introduce here the first public-access dataset of thermal imaging of alcohol-impaired subjects captured in Vietnam. The database contains 2,280 thermal facial images of 66 volunteers (in equal numbers of sober and intoxicated), captured using a Total Meter HT-03 thermal camera with high resolution and obtained in diverse environmental conditions ranging from outdoor to indoor with temperatures ranging from 22 to 37 degrees Celsius. Each subject was captured both sober and intoxicated, with as much diversity in light, camera positions, and time as possible. This is a very prominent characteristic of this data concerning other publicly available data for intoxication in the world. All data are annotated with binary class labels (sober/intoxicated) and are meant to aid research in computer vision, biomedical signal processing, and public safety. This information can be used to support AI-based non-contact intoxication detection technology and serve as a standard for environmental and cross-cultural alcohol-related thermal analysis research.