Uma ferramenta para demarcação de áreas de interesse e monitoramento em tempo real de pessoas usando aprendizado profundo

Abstract

The identification of people is a relevant problem in the area of Object Detection. More recently, techniques have emerged to perform these detections in real time, using Neural Networks and Deep Learning. These techniques have the great advantage of being able to quickly and accurately classify persons in a scene, which are indispensable properties in video monitoring applications. Deep Learning based solutions have a wide range of applications. Since Neural Networks began to be successfully applied in the area of Supervised Learning, approaches that make use of them are achieving state-of-the-art results in solving classification problems (such as speech recognition, remote sensing, among others). These technologies can be combined with other techniques to solve real-world geometric problems, such as tracking a user’s location using virtual reality glasses. In this work, we report the process of developing a tool that detects people and infers whether they are in a demarcated area or not. This program aims to enable real-time monitoring of regions customized by the user, and can be used for various purposes. The system was implemented in the Python programming language, which provide practicality and also has a vast usability of libraries, methods and functions that are currently used in the area of Computer Vision. We also use a pre-trained Yolov5 model to detect people more confidently and quickly.