APPLICATION OF NEURAL NETWORKS TO DETERMINE THE MAIN CHARACTERISTICS OF TRAFFIC FLOWS IN THE CITY

Abstract

A method for automatic vehicle recognition and classification is proposed, based on the Mask R-CNN convolutional neural network. The developed method enables automated retrieval of information on the composition and intensity of traffic flows for each vehicle type, which is the basis for any method for calculating pollutant emissions from motor vehicles. The article describes the stages of neural network training, as well as the results obtained from its use. The developed method for assessing traffic flow intensity, based on the analysis of data obtained from video cameras installed on highways, using the Mask R-CNN convolutional neural network, has demonstrated good results. By expanding the test sample, the number of recognized vehicle classes can be increased.