Traffic Control System Development Based on Computer Vision

Abstract

In 2020, the number of Brazil inhabitants was approximately 212 millions, whereas the number of vehicles jumped from 104 to 107 millions. It is observed, therefore, that there is 1 vehicle for every 2 inhabitants. That same year, the population in urban areas went to 80%. Managing traffic in big cities is becoming a huge challenge. Traffic lights operating with Fixed Time Signal to control vehicle flux are no longer efficient in all traffic scenarios. Technological advances in Computer Vision, moving objects detection and classification techniques and the demanding of little computer power to manage these tasks allowed the development of TEOP, a CV-based traffic control system. This low cost solution was implemented to advance the Fixed Time Signal system, cameras and logical network infrastructure already in use in Brazilian cities, transforming regular traffic lights into smart traffic lights (STL). An application was developed and installed in a computer to capture images of traffic, count vehicles and calculate time needed for them to pass through. Raspberry controlled traffic lights. In comparison to regular traffic lights, STLs improved traffic flow by 33%, allowing a wait of just 4 s in cases where there is no vehicle flow at the competing traffic light and reducing the crossing time of 46 vehicles from 152 to 116 s when there were 12 vehicles on the competing side, a significant gain. It was also capable of reporting traffic jams and creating a database that could be used for decision-taking by agencies responsible for each route.