Análise de veículos em cruzamentos com semáforos utilizando Deep Learning

Abstract

Traffic lights have been used since the 20th century in many countrys of the world to control street traffic, but with the number of vehicles increasing dramatically, there are many issues with the massive increase of car ow in the streets and avenues, which can cause kilometric congestions. An intelligent and e cient solution should be developed to solve these problems, and in addition, optimize travel and waiting times at street crossings. This work presents a simple implementation of an intelligent traffic light system which receives a video stream and processes it using digital imaging processing to detect and count vehicles, in order to control the ow at real time. With this system, tests were made using images from a traffic light perspective, simulating a real traffic camera scenario, to analyze the feasibility of using this system, based on the number of correct detection results of vehicles. The results aqurided here, showned that, from four scenarios tested, the system obtained success in three of them, and the any main error was because of a car stopped behind a bus, making it very di cult to detect. Furthermore, this system can be used in traffic lights to detect and count vehicles, serving as idea for related works which permits a better control of the traffic ow.