MAPEAMENTO DE POSTES POR IMAGENS RGB DE NÍVEL DE RUA UTILIZANDO MÉTODOS DE DETECÇÃO DE OBJETOS: UM ESTUDO DE CASO NA CIDADE DE CAMPO GRANDE, MS

Abstract

This dissertation aims to develop and apply robust methods for the detection and classification of objects, with a specific focus on electrical poles present in RGB images of the electrical grid collected in the city of Campo Grande, Mato Grosso do Sul, Brazil. The central objective is to demonstrate the utility of these methods in the context of the electrical energy sector, highlighting their capability to identify maintenance needs. The accurate identification of electrical poles in images of the electrical grid plays a critical role in enhancing the electrical energy sector. This efficient identification is essential for evaluating maintenance demands and detecting structural faults, thereby contributing to the improvement of infrastructure management. The adopted methodology encompasses the construction of an annotated image dataset, where electrical poles and other elements of the electrical grid are manually identified and delineated. These annotated images are used to train object detection methods, encompassing techniques such as machine learning algorithms and convolutional neural networks. Through experiments, these methods are evaluated and compared in terms of accuracy, processing speed, and generalization. The conducted experiments yield results that underscore the effectiveness of the developed methods for the detection and classification of electrical poles in RGB images of the electrical grid. Comparative analysis of the tested models enables the identification of the most suitable method for the application at hand. With these contributions, the dissertation seeks to optimize infrastructure management, enhancing maintenance processes, and improving the efficiency of energy supply. Keywords: Electrical Energy, Electrical Grid, Object Classification, Object Detection, RGB Images.