ÀÖÓãºǫ́

The Dutch police will start using traffic enforcement cameras to automatically recognise drivers who are holding a mobile phone. The technology behind the pilot has been developed with Utrecht University. Computer science researcher Ronald Poppe supervised the pilot project.

After years of decline, the number of traffic accidents in the Netherlands has recently been on the rise again. This trend is partially attributed to the use of mobile phones behind the wheel. Using a phone while driving a car is strictly forbidden, but enforcing this ban is time-consuming. Currently, drivers have to be caught and stopped by police officers. In practice, this means that only a few offenders per hour can be fined. As a result, the chance of being caught is low.

Technical solution

The police wanted to increase this chance, and were interested in the possibilities of computer vision: the automatic analysis of images by computers. In early 2018, Computer Science student Jannes Elings started his graduation project with the Dutch police. The aim was to develop a method to automatically analyse camera images to see whether a driver is operating a mobile phone. "A challenging task," says supervisor Ronald Poppe, "especially because people sit in their cars in very different ways. We saw people holding phones on their laps, and in addition, it turns out that people hold all kinds of objects in their hands or against their ears."

Ultimately, the researchers decided to identify several important components in the image: the driver's head and hands, a car kit and the cellphone. Deep learning, a form of computer vision in which complex models are learned from large amounts of data, was used for this purpose. By cleverly looking at where the phone, hands and head are in relation to each other in the image, a distinction can be made between safe and unsafe driving behaviour. A test has shown that this approach misses few offenders and does not cause a false alarm all too often. It also turned out that the quality of the method could be improved still further by presenting more examples to the computer.

Last hurdle

By now, that is exactly what the Dutch police have done. The initial prototype has been further developed and is now ready for use. The last hurdle to be overcome is legislation, which still requires a driver to be stopped. As soon as that is complete, the Dutch roads will be made a little safer again.