DRIVE: In-Vehicle Drunk Driving Detection
Driving under the influence of alcohol is among the most relevant reasons for deaths or severe injuries in public road traffic. The main problem with driving under the influence of alcohol is the significantly decreased driving performance. Over the past decades, several measures helped to reduce the amount of traffic accidents related to alcohol. The stagnating number of alcohol-related accidents in recent years show that existing measures have reached a saturation in their effectiveness. However, the use of in-vehicle sensor data streams (e.g., real time steering data, driver camera-based eye tracking) in conjunction with machine learning promises an effective path in the detection and prevention of drunk driving that is cost-effective, scalable, and reaches drivers in-situ.
The overall project goal of DRIVE is to build a reliable in-vehicle drunk driving detection system and targets the following two main research questions: (1) To which degree of accuracy can drunk driving be detected from today’s real-time vehicle sensor data streams, i.e., Controller Area Network (CAN-bus) data? (2) To which degree of accuracy can drunk driving be detected from future real-time vehicle sensor data streams, i.e., driver video monitoring (incl. eye tracking) and physiological data?
In the short-term DRIVE has the potential to increase road safety and reduce accidents by detecting drunk driving and providing corresponding driver warnings. In the long term, we deem DRIVE as a first step towards a scalable system for detecting general driver impairment. Based on safety-relevant deviations from their normal driving behavior, such a system could inform drivers in a timely manner about their limited ability to drive. No matter if they are impaired by drugs or by medical incapacitations, e.g., hypoglycemia in case of people with diabetes.
DRIVE is a collaboration of the University of Berne (Institute of Forensic Medicine) as well as the Bosch IoT Lab and Center for Digital Health Interventions (CDHI) at ETH Zurich and the University of St. Gallen.