Mathematical support for monitoring pollutant emissions from vehicles in the regulated intersection area based on neural network algorithms

In modern cities, environmental issues related to motor transport occupy an increasingly important place in the management system of urban transport flows. Vehicles emit the greatest amount of exhaust gases during sudden changes in traffic modes, which is typical of signalized intersections. Traffic congestion situations also occur more often at intersections, when a large concentration of traffic creates an unfavorable environmental background. To obtain operational information about the parameters of the intensity of traffic flows, the authors used neural network algorithms for recognizing vehicles from video streams received from stationary street surveillance cameras at city intersections. An optimized algorithm for the operation of a trained neural network (YOLOv4) allows for extracting and interpreting data on traffic flow parameters in real time. As part of the study, the developed mathematical models made it possible to implement real-time monitoring of the amount and concentration of pollutants from vehicles in the controlled intersection zone. The calculation of the amount of pollutants released into the atmosphere from transport is carried out taking into account the average speed, type of vehicle and idle time in the measurement area. An appropriate software system can serve as a basis for predicting the difficulty levels and environmental risks of various atypical traffic situations.