Drivers will believe in special signs – Auto – Kommersant

United Operator LLC (part of the Avtodor Group of Companies) is starting to test a neural network with the ability to read dirty or unreadable car numbers on the Central Ring Road and M-12 Vostok. Their owners do not pay tolls and avoid fines. The neural network will analyze the database of all vehicles that have ever passed through the toll sections in search of suitable matches. In the case of a “positive result” of the experiment, the company will propose changing the legislation to impose fines based on the testimony of the neural network. Determining a car by body type, configuration color, dents cannot guarantee 100% identification, experts warn.

The United Operator company announced plans to test software with “elements” of a neural network capable of recognizing dirty or unreadable vehicle numbers. The system will be launched in pilot mode at the beginning of 2023 at the Central Ring Road and at the open sections of M-12 Vostok.

Let us remind you that the toll roads being put into operation work according to the Free flow system: there are no barriers on the highway, the passage of cars is recorded by sensors and cameras. The fee is debited from the transponder’s account. If there is no transponder, the camera will recognize the number in order to bill the car owner later. The fine for violators is 1.5–5 thousand rubles. The system is used on the Central Ring Road and open sections of the M-12 Vostok, it is planned to be introduced on the M-4 Don.

At the end of 2021, the collection rate at the Central Ring Road was 95%, in the summer of 2022 the figure began to decline (there are no exact data at the moment). Among the non-payers there are also owners of vehicles whose license plates could not be recognized by the cameras. As explained by “Kommersant” in the “Single Operator”, the software solution will “recognize” dirty or unreadable numbers. “A neural network based on machine vision will analyze the entire database of toll roads of the Avtodor state company,” the company said. numbers have been read.

In the case of a “positive result” of the experiment, the company will propose changing the legislation to impose fines based on neural network data.

“In the future, the technology can also be trained to detect deliberately hidden numbers based on the distinctive features of cars, such as dimensions, scratches, plates, stickers, and others,” said Marcel Nigmetzyanov, General Director of Single Operator.

The state company Avtodor yesterday could not provide a response to Kommersant’s request.

The capabilities of the neural network, we recall, are already used in regional systems for fixing traffic violations. In Moscow, drivers talking on the phone while driving are identified in this way, in the Moscow region – taxis with violations of the rules for coloring.

“If the vehicle does not have pronounced visual individual differences from the factory configuration, the risk increases when non-payment and a fine will be imputed to the car owner on a model similar to the violator (with the same coloring and equipment),” says Grigory Shukhman, an expert on traffic cameras.— In the end, Tom will have to prove that he did not violate. Claims will be made on the basis of the argument “your car is similar to the violator’s car that we recorded.” Determining a car by body type, color, configuration, dents cannot guarantee 100% identification, especially on dirty winter roads, Serhiy Radko, a lawyer for the Freedom of Choice movement, confirms: state registration can not be the basis for punishment.

Ivan Buranov