Authors:

Eya MLIKA 1, Bahe HACHEM 2, Yamen AL HABASHI 2, Loic DEGUELDRE 2, Luc DUONG 1

1 École de Technologie Supérieure, Montréal QC, Canada;
2 Numalogics, Montréal QC, Canada

DOI:

https://doi.org/10.15221/24.23

Full paper:

PDF

Keywords:

Respiratory masks, Customization, Artificial Intelligence, Face Mask, 3D Face Reconstruction, Mesh generation, non-rigid deformation, ARKit, Deep Learning, Personal Protective Equipment, Safety, Health Care Safety, Industrial Safety, Facial structure interaction, thin plate spline, TPSNet

Abstract:

Respiratory masks are highly used in healthcare and industrial environments to protect individuals from airborne contaminants and infectious pathogens. However, conventional respiratory masks are often one-sized, therefore it becomes challenging to fit diverse facial morphologies. If loosely fit, the mask may not adequately safeguard or be over-tight to achieve protection, which could cause discomfort or even generate pressure wounds over extended wear.
This project aims to propose an artificial intelligence approach for the design and customization of respiratory masks, emphasizing customized products to better fit into the variety of human face morphology. The proposed approach begins with the creation of facial geometries using 3D facial data obtained by using the ARKit framework. ARKit allows to acquire a structured and complete mesh of the subject's face despites incomplete and noisy data. Each facial scan captures 5,023 3D points, providing a detailed map of individual facial features. The resulting dataset, including 60 different facial scans, forms the basis of our machine learning algorithm. This algorithm is designed to improve the customization and fit of respiratory masks, enhancing wearer safety and comfort. From this facial modelling, a deep learning model designed to predict the deformations of a mask when fitted to the face was deployed. The model enables the identification of potential areas of pressure and mask misfits, predicting problems before they become critical. A predictive model was further introduced to simulate the interaction between the facial structure and the mask as closely as possible. Combining scanning technologies and predictive modelling will alleviate the detection of gaps and pressure points, enabling preventive measures to be taken to rectify these defects. Due to an in-depth understanding of these interactions, the newly developed model proposes modifications to the mask design to better match the unique contours of each face, thus improving the mask's seal and comfort.
This research might contribute to improve fitting and address important health protection issues and accelerate safety regulation compliance, thereby lessening health risks related to the long-term use of poorly fitted masks. Adequate fitting might have an important impact on the design of personalized protective equipment.

How to Cite (MLA):

Mlika, Eya et al., "Personalized Fitting of Respiratory Mask Using Deep Learning and 3D Facial Modeling", 3DBODY.TECH Journal, vol. 1, Oct. 2024, #23, https://doi.org/10.15221/24.23.

Presentation:

VIDEO availble in proceedings

Details:

Volume/Issue: 3DBODY.TECH Journal - Vol. 1, 2024
Paper: #23
Published: 2024/10/30
Presented at: 3DBODY.TECH 2024, 22-23 Oct. 2024, Lugano, Switzerland
Proceedings: 3DBODY.TECH 2024 Proceedings

License/Copyright notice

Copyright © 2024 by the author(s).
This work is licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
The papers appearing in the journal reflect the author's opinions. Their inclusion in the volumes does not necessary constitute endorsement by the editor or by the publisher.

Note: click the + on the top left of the page to open/close the menu.