Unified framework for automatic vehicle recognition and classification for electronic toll collection systems in the Free-Flow model

• Grant number: 23/13235-8
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: November 01, 2024
• End date: July 31, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: Jorge Alberto Mathias de Almeida
• Grantee: Jorge Alberto Mathias de Almeida
• Company: ADNEW TECH SOLUCOES EM TECNOLOGIA DA INFORMACAO LTDA
• CNAE: Desenvolvimento de programas de computador sob encomenda ; Desenvolvimento e licenciamento de programas de computador customizáveis ; Tratamento de dados, provedores de serviços de aplicação e serviços de hospedagem na internet
• City: São Paulo
• Associated researchers: Luiz Gomes Trevisani
• Associated scholarship(s): 24/20779-7 - Comparative study of image processing and machine learning techniques for automatic vehicle recognition in toll collection systems., BP.TT

Abstract

Right of way fees on roads and highways in Brazil are often collected through toll plazas. This system consists of specific points on the road where vehicles are obliged to stop so that the fare can be collected. To make the process easier and reduce queues, there is the option of using an automatic payment system, which requires the prior contracting of the service through a Collection Service Operator (OSA) and the installation of a radio frequency device in the vehicle. This device identifies the user, carries out the transaction and authorizes the passage of the vehicle, allowing the use of exclusive lanes intended for vehicles that have the identification tag. However, the radio frequency identification system has limitations related to cost, environmental and electromagnetic interference, as well as security and privacy issues. In contrast, many countries have adopted electronic toll collection systems that allow collection through automatic identification and classification of vehicles, without the need to stop at collection points. This system, known as the "free-flow system", has numerous advantages, but it also has technological limitations that have stimulated the continuous development of solutions to improve it and make it more accessible and reliable for users. Brazil, as a developing country and considering the novelty of this model in the country, faces challenges in implementing the electronic toll collection system. It is therefore crucial to carry out detailed studies to assess the suitability of existing systems (all foreign) to the Brazilian reality and to develop more suitable and possibly more economical national solutions. The initial costs of implementing such a system are high and it is also important to consider the ongoing costs of maintaining, updating and providing technical support for the equipment. Dependence on international technology can lead to high maintenance costs for imported solutions. Another problem to be tackled in Brazil is toll evasion, which results in significant losses for the concessionaires responsible for managing the country's highways. Given this scenario, the overall aim of this project is to develop a computational framework based on deep learning algorithms, such as convolutional neural networks with darknet deep learning architecture, to enable automatic recognition and classification of vehicles based on images, in the context of a free-flow electronic toll collection system. Specifically, automatic recognition will be based on images of vehicle license plates, including the stages of localization, detection and segmentation of license plate characters, as well as optical character recognition (OCR). The framework will also be able to classify vehicles according to their shape, size and other characteristics, making it possible to determine the total number of axles and the number of suspended axles (in cargo vehicles), factors that influence the toll rate, as well as identifying exempt vehicles. In order to validate the framework, exhaustive tests will be carried out in the laboratory, using images obtained under real conditions from public databases and from a bank of images obtained from controlled gantry cranes, through collaboration with the CCR concessionaire. This approach aims to ensure the robustness and applicability of the framework in practical and controlled situations. By combining advanced vehicle recognition and classification techniques, this project offers a promising alternative to overcome the limitations of current systems, optimizing road management and providing a more agile and accurate experience for users. (AU)