Researchers at Edith Cowan University (ECU) in Australia may have developed a breakthrough technology that could help save lives on the roads. They have developed a new computer algorithm that can determine if a driver has been drinking while driving. This AI technology could help police detect drunk drivers sooner than they currently can, preventing deadly accidents before they happen.
In the future, such technology may be implemented in smarter cars, preventing drunk drivers from actually starting to drive while drinking.
The researchers published their work and presented their findings at the IEEE/CVF Computer Vision Applications Winter Conference.
Australian scientists partnered with Mix by Powerfleet to collect video data of drivers under the influence of alcohol observed in a controlled, realistic environment. Before using a simulator to capture camera footage, the researchers split participants into three groups: sober, slightly intoxicated, and severely intoxicated.
The researchers fed the video into a machine learning system (AI), which interpreted changes in standard RGB video of the driver's face to determine the level of drunkenness. The AI analyzed various aspects, including facial features, gaze detection, and head position.
“Our system detects different levels of alcoholism disorder with an overall accuracy of 75 percent across three levels of classification,” ECU doctoral student Ensier Keshtkaran said in a statement.
While this rate may not be high enough, it's still a great achievement, and if this technology can be used to catch three out of four drunk drivers early, it could have a huge impact on preventing accidents.
According to the ECU, drink driving is the primary cause of approximately 30% of fatal accidents in Australia, and statistics show that one in five drivers killed on Australian roads had a blood alcohol level of 0.05 or above.
The researchers can always improve their AI model with additional experiments, and they are already trying to determine what kind of image resolution the AI needs to detect drunk drivers.
If the technology is integrated with roadside surveillance cameras, police will be able to prevent drunk driving more effectively than by conducting random tests or stopping drivers who are driving erratically.
This is the system's main advantage: it is a novel approach that could allow police to catch drunk drivers at the start of their journey, rather than after they have been driving a vehicle for an extended period of time under the influence of alcohol.
Future versions of the system could be integrated into road cameras that can already detect other types of risky driving behavior, such as cell phone use or not wearing a seat belt.
Keshtkaran also mentioned efforts to incorporate driver alcohol detection systems in future cars. Some may focus on algorithms that observe driving behaviors related to real-world driving. Steering patterns, pedal usage, and speed can be indicators of drunk driving. Alcohol detection sensors can also determine if a person is planning to drink and drive.
However, current systems also do not consider using computer vision techniques to detect the risk of drunk driving.
ECU researchers also note that other detection methods require the driver to start driving so that the algorithm can evaluate their behavior. By the time the system renders its decision, the drunk driver is already a danger on the roads. Using AI to interpret real-time footage may be a faster, safer way to detect drunk driving before a fatal accident occurs.
It's still early days though, and more research may be needed before Australian authorities are willing to incorporate such technology into road safety cameras and change the laws to allow the cameras to be used for this purpose. But this is certainly an exciting way to use AI to potentially prevent drunk driving accidents, and one that authorities and smart car vendors around the world should consider.
