Visão computacional: estudo comparativo de algoritmos de subtração de fundo aplicados em soluções para o gerenciamento de tráfego urbano de veículos e pedestres

Abstract

The growth of the population and the fleet of vehicles all over the world, particularly in the big cities, have caused complications in terms of traffic. Solutions for reducing congestion and improving vehicle trafficability require effort and time, which is expensive and ineffective in the medium and long term considering Brazilian municipal administrations. This situation stems mainly from the need for adequate methodologies for efficient monitoring and data collection that public managers can analyze to serve as a basis for road and traffic infrastructure policies. Intelligent traffic monitoring is a new branch of intelligent transport systems that focuses on improving urban traffic conditions. The main objective of systems based on intelligent monitoring is to improve the traffic system, reduce vehicle fluidity problems and improve urban mobility. Many Brazilian cities face congestion due to the inefficient configuration of traffic lights, mainly based on fixed-cycle protocols. Within this context, there is a need to improve and automate the existing traffic light system. A mixed technique of artificial intelligence and computer vision may be desirable to develop a reliable and scalable traffic system that can help solve these problems. It first works in computer vision date back to the late 1970s, when computers could already process large data sets, such as images. In-depth studies in this area begin, and the field of computer vision can be characterized as immature and diverse, even after there are recognized works. Within this context, this dissertation focused on using computer vision technology to build an efficient and resourceful system for synchronous and automated traffic analysis. The performance of different algorithms and systems that apply computer vision to identify and track objects in urban traffic is discussed, detecting pedestrians and vehicles. With this, determining which algorithms and systems would best fit a more complex system based on IoT devices or even autonomous vehicles, which use computer vision systems. Somebody, Systematic Literature Review (SLR) were carried out in articles published from 01/01/1970 to 01/31/2023. As BGS is a technique used in computer vision to detect moving objects in a sequence of images obtained from a video, for example, it was found that relying only on BGS without any other processing or machine learning (ML) support or deep learning (DL), are not accurate enough for direct application in traffic light systems or autonomous monitoring. However, with the constant evolution of specialized software and hardware, the joint application of these solutions is promising. As the results of this work demonstrate, there is a need to improve the individual filters and, in this way, try to make them more reliable when applicable and implement them aided by other technologies, in an attempt to become more assertive in terms of precision and other metrics, such as false positives and true positives.