Name: CÉZAR AUGUSTO GOBBO PASSAMANI
Publication date: 16/10/2024
Examining board:
Name |
Role |
|---|---|
| ALBERTO FERREIRA DE SOUZA | Presidente |
| ANDRE GEORGHTON CARDOSO PACHECO | Examinador Interno |
| CLAUDINE SANTOS BADUE | Coorientador |
| JURANDY GOMES DE ALMEIDA JUNIOR | Examinador Externo |
Summary: This thesis presents an innovative method for assessing the risk of COVID-19 contamina-
tion, focusing on social distancing and face masks. The proposed system leverages images cap-
tured by surveillance cameras to compute a real-time health risk indicator in various public
spaces, including squares, streets, restaurants, and shopping centers. The system analyzes the
number of individuals wearing masks and their proximity to each other, providing an accurate
assessment of COVID-19 contamination risk. Our approach employs deep neural networks for
detecting individuals with and without masks alongside advanced computer vision techniques
to measure the distance between them. The system addresses several challenges, such as distin-
guishing face masks at varying distances and angles relative to the camera, dealing with occlu-
sions, and managing face masks' diverse shapes and sizes. To overcome these hurdles, we de-
veloped and released a comprehensive face mask detection dataset containing 44,402 images
of faces captured in varied and challenging scenarios. This dataset was used to train our neural
networks to high-performance levels. Our best deep neural network architecture achieved good
results through rigorous training and validation, with 91.41% precision, 82.88% accuracy,
and 89.88% recall in face mask detection. These results demonstrate the system's effectiveness
in mitigating the risk of COVID-19 contamination in public environments. Our system provides
a valuable tool for public health authorities, enabling real-time automated health monitoring
and laying the groundwork for future research and advancements in this field.
