In six months, only in districts near Moscow, traffic enforcement cameras recorded 154,000 cases when the driver had not fastened the seat belt.
Simple arithmetic suggests that nationwide statistics would show an eight-digit number of violators for the past half of the year and more than a million each month.
Meanwhile, according to the WHO, fastened seat belts reduce the likelihood of death for the driver and the front passenger by 45-50%, and by 25% for the second-row passengers.
The use of cameras for detecting unfastened seatbelts began in 2020, when the first units capable of recognizing this violation became available. Today, according to the Research Center for Road Traffic Safety of the Russian Ministry of Internal Affairs, more than half of the 27,000 installed camera units can detect an unfastened seat belt.
However, the offenders are just beginning to learn about this innovation, as the system was first tested and trained in several regions and is under way to reach full capacity.
At first, an unfastened seat belt is identified by a neural network. Moreover, traffic violation cameras in the Moscow region are trained to see the seat belt zone, where it is attached to the car body behind the driver's shoulder – if the belt is fastened, the neural network will notice it for sure.
The selected records are then handed over to the traffic police inspector, who makes the final decision to issue a traffic ticket. Thanks to ITS, the traffic police receives more photos and videos of the car, so the inspector has a much better opportunity to look at them in detail. In addition, the resolution of this photographic evidence is higher than that of the printed traffic violation notice. The inspector can zoom in on the image without any loss in quality and decide whether a traffic violation has taken place.
To make it function correctly, the neural network was trained on half a million cases in order to eliminate possible errors. However, tickets for unfastened seat belts are sometimes issued in error. It is obvious that neural networks are still not quite perfect, but their quality and capabilities are constantly increasing.