Desenvolvimento de um modelo para identificar, categorizar e mapear dados precisos da qualidade de placas de trânsito de vias públicas através de câmeras veiculares

Abstract

The recognition of traffic signs is one of the many utilities enabled by embedded systems. Through the use of vehicle cameras, it is possible to capture and classify traffic signs with the aid of Artificial Intelligence, specifically Convolutional Neural Networks. In addition to addressing the basic problem solved in the literature, which consists of detecting and classifying the type of traffic sign, the present project achieved a methodology for creating databases and models to evaluate the quality of the signs, which results in the possibility of mapping this data through the implementation of such a system. With this, it is possible to improve the management of traffic sign maintenance carried out by the government or contracted agency. The main results obtained were a dataset with 7805 images and 17493 Brazilian plates, as well as object detection and classification models trained with this data and validated with good performance. Another important result of this research is the application and analysis of data augmentation techniques, which proved capable of increasing the robustness of the models to situations where the collected data is not sufficient.