ORUMA, SAMSON OGHENEOVO and Covenant University, Theses (2021) DEVELOPMENT OF A LIGHTWEIGHT MODEL FOR COVID-19 FACE MASK WEARING-POSITION DETECTION. Masters thesis, Covenant University.
![[img]](http://eprints.covenantuniversity.edu.ng/style/images/fileicons/application_pdf.png) |
PDF
Download (186kB) |
Abstract
The Corona Virus Disease (Covid-19) spread has led to many infection cases with several resulting deaths. The increasing number of new Covid-19 variants has reinforced the need to develop a proactive critical preparedness, readiness response action plan. This study aims to develop a lightweight model for detecting face mask-wearing positions using a locally generated dataset of black people. A six-fold methodology of dataset generation, data preprocessing, model selection, model training and validation, and model deployment was adopted for this study. A dataset of black people from three universities in Nigeria was generated for males and females, indicating four distinct face mask-wearing positions and eight classes, using a digital camera, smartphones and medical face mask. The images were subject to preprocessing such as cropping, resizing, labelling and data augmentation. The generated dataset was used to train a modified YOLOv5s model and deployed using Roboflow’s webcam platform and local PC with Pycharm IDE. The developed model achieved 94.2%mAP, 94% recall, and 79% precision on Roboflow’s platform after training for 250 epochs. Training on Google Colab platform for 100 epochs resulted in 91.5% mAP, 91.4% recall and a precision of 86.8%. A dataset called "Black Face Mask Dataset" was generated from this study, with 13 different annotation formats. This study’s outcome will be beneficial to researchers in computer vision and the government of developing countries. The generated dataset can be merged with the existing face mask detection dataset to achieve a better model with good black people representation. the eight classes can be merged into smaller classes based on the application requirement to produce higher object numbers per class. The developed model can be cloned from GitHub for easy integration without the need for retraining.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Object detection, Image classification, Covid-19 prevention, Face mask detection, Black-coloured people, Deeplearning. |
| Subjects: | T Technology > T Technology (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | Mrs Hannah Akinwumi |
| Date Deposited: | 29 Nov 2021 14:35 |
| Last Modified: | 29 Nov 2021 14:35 |
| URI: | http://eprints.covenantuniversity.edu.ng/id/eprint/15522 |
Actions (login required)
 |
View Item |
