In a significant advancement for road safety, researchers from Centre for AI&ML, Edith Cowan University, Western Australia and MiX Telematics, Western Australia have developed an innovative artificial intelligence system capable of detecting drunk driving with remarkable accuracy. This groundbreaking technology, detailed in a paper published on April 9 as part of an IEEE and CVF conference, uses standard commercial RGB cameras to predict critical levels of blood alcohol concentration (BAC) in drivers.
The system, which boasts a 75% accuracy rate, can detect alcohol intoxication impairment as subtle as 0.05 g/dL – the WHO recommended legal limit for driving. Unlike existing methods that rely on observable driving behaviors, this new approach can assess a driver’s intoxication level immediately upon entering the vehicle.
Using a single color camera, the AI analyzes facial cues such as gaze direction and head position. The system can also incorporate 3D and infrared footage of the driver’s face, rearview videos showing posture, and data on steering interactions and driving behavior.
To develop and test this technology, researchers compiled a dataset of 60 subjects engaged in simulated driving scenarios across three levels of alcohol intoxication: sober, low, and severe. They collaborated with software company MiX by Powerfleet to collect data from alcohol-impaired drivers in controlled but realistic environments.
The algorithm successfully identified visual indicators of intoxication, including bloodshot eyes, flushed face, droopy eyelids, and a dazed look. This project represents the first large-scale real-life dataset of alcohol intoxication used to assess intoxication levels using off-the-shelf RGB cameras for drunk driving detection.
As drunk driving continues to be a significant contributor to global road injuries, this AI system holds great promise for improving road safety. By providing real-time monitoring of a driver’s BAC, it could potentially prevent alcohol-related accidents and save countless lives.
