The Use of the YOLO neural network in the task of separating vehicles and pedestrians at a signal-controlled intersection

Abstract

© 2020 IEEE. Many studies deal with the problem of ensuring safety and increasing the capacity of signal-controlled junctions of the road network (RN). The use of innovative developments in the field of computer vision to model intelligent transport systems (ITS) has been a steady trend in the framework of the development of smart cities. One of the less-studied issues is the optimal respecting the interests of drivers and pedestrians within pedestrian crossings when turning right. In our work, we focused on the problem of ensuring the minimum time of waiting and crossing signal-controlled intersections by pedestrians. We proposed to solve this problem based on adaptive traffic light control through the use of computer vision. To solve the problem, we trained two configurations of YOLOv3 neural networks based on the Darknet framework with additional modifications. The SORT tracker was used to track objects. In the zones adjacent to pedestrian crossings, we used a neural network to search for pedestrians intending to cross a section of the carriageway, which conflicts with turning vehicles. Based on the data on the presence of pedestrians and the information on the composition of the vehicle queue, we calculated the optimal time-sharing of the permissive traffic light interval.