A Study of the Impact of the Transport Queue Structure on the Traffic Capacity of a Signalized Intersection Using Neural Networks

Abstract

The article deals with the development of a computer system, which allows us to recognize vehicles, track them, and measure the time needed to cross an intersection by each car in the lane. The main area of research is the analysis of the dependence of the intersection crossing time on the position of vehicles in the queue formed in the traffic lane. To count the vehicles in the queue and determine their category, we used the Yolo v3 neural network and the SORT tracker modified to return the object class. The article describes in detail the proposed algorithm for collecting the data on the queue of vehicles: the number of vehicles in the queue and their classes, the time of passing the stop line and crossing the intersection, as well as determining the driving direction. All vehicles are divided into three categories depending on their acceleration. We analyzed the collected data on the queue structure and the time of its unloading and demonstrated their direct interconnection.