Uma Análise Do Yolo8: Detecção De Pessoas Com Máscaras Faciais

Abstract

The research aimed to apply the Yolo method (you only look once) to detect specific objects in different images, since computer vision advances rapidly and over the years performance improvements are made so that it can be used in the area of health or construction, for example. In order to carry out the research, it was necessary to have a theoretical basis on this technology and its various functions and operations, such as: classification, detection, segmentation, tracking and pose through the github of the company "ultralytics" which developed the technology. In the research, detection was used to distinguish between people who are and are not using face masks. The obtained results were considered evaluating two metrics: Map50 and time in two types of GPUs: A100 and V100. In the results it was observed that using the GPU V100 had an average Map50 of 0.862 and average training time of the model of 56.77 minutes, while the GPU A100 had an average Map50 of 0.866 and average training time of the model 13.06 minute model, both results for 100 epochs.