Videomonitoramento em Espaços Inteligentes: Uma Aplicação de Segurança Pública em Identificação de Armas de Fogo

Abstract

The amount of crimes involving firearms is growing in different parts of the world. One way to increase security in society is through automating surveillance systems. In this context, an intelligent video surveillance system is a viable solution to be implemented in order to collaborate with other public safety actions. In this sense, an intelligent space was set up in the laboratory of Ifes - Campus Guarapari, formed by video surveillance cameras and computers, with the objective of recognizing people carrying firearms. In addition, a database containing approximately nine thousand images was obtained. This paper aims to present this intelligent environment with video surveillance that, through the deep learning model YOLOv5, recognizes firearms within the environment. Tests were performed with the own dataset and with a public dataset. The precision values recall and mAP (Mean Average Precision) were used to evaluate the results. The system proved to be robust to partial weapon occlusion, rotation and scaling. The results obtained showed that the proposed system achieved weapon recognition with up to 91.39% accuracy, 89.91% recall and 88.78% mAP. In this way, the proposed intelligent space shows a viable alternative to be studied and applied on a larger scale in environments with the objective of reducing the amount of crimes involving firearms.