Face Mask Detection Using Deep Learning

Sumit, Sahil Bhasin, Parv Patel, Princess Singh

Abstract
accurately detecting face masks
This project aims to develop a face on individuals.
mask detection system utilizing deep • To evaluate the system's
learning techniques to mitigate the effectiveness in various
spread of infectious diseases. By environments and lighting
employing Convolutional Neural conditions.
Networks (CNNs), the system will be • To implement a prototype that
capable of identifying individuals can be integrated with existing
wearing face masks and those not surveillance systems for public
complying. This proposal outlines the health monitoring.
problem statement, objectives,
3. Methodology
methodology, expected outcomes,
and potential implications of 3.1 Data Collection and Preparation
implementing such a system in public
A diverse dataset containing images
spaces.
of individuals with and without face
masks in various environments will be
compiled from public sources and
1. Motivation
potentially augmented to increase its
Despite the known effectiveness of variability. This dataset will be
face masks in preventing disease annotated with labels indicating the
spread, compliance remains presence of a face mask. Data
inconsistent. Manual enforcement of augmentation techniques such as
mask policies is labor-intensive and rotation, scaling, and flipping will be
not always effective. A technological applied to improve the robustness of
solution could provide consistent and the model.
unbiased enforcement, supporting
3.2 Model Development and
public health efforts.
Architecture
2. Objectives
The project will employ Convolutional
• To design and develop a deep Neural Networks (CNNs) due to their
learning model capable of proven effectiveness in image
recognition tasks. A comparative
study of existing architectures (e.g., 4.Experiments
ResNet, MobileNet) will be conducted
4.1 Performance Evaluation
to select the most suitable model for
processing and high accuracy. The Experiments will be conducted to
selected model will be customized evaluate the model's accuracy in
and optimized for the specific task of detecting face masks under various
face mask detection. scenarios, including different lighting
conditions, mask types, and angles of
3.3 Training and Validation
the face. The model's processing time
The model will be trained using the will also be measured to ensure its
prepared dataset, utilizing techniques suitability for applications.
such as cross-validation to assess its
4.2 Comparative Analysis
performance. Training will be
performed on a GPU-enabled A comparative analysis will be
platform to accelerate the learning performed against existing face mask
process. Performance metrics detection systems to benchmark our
(accuracy, precision, recall, and F1 model's performance. This analysis
score) will be used to evaluate the will highlight the advantages of our
model. Techniques like transfer approach in terms of accuracy, speed,
learning may be employed to and operational efficiency.
leverage pre-trained models for 4.3 Testing
enhanced efficiency and accuracy.
The prototype will undergo testing in
3.4 Deployment and Integration scenarios to validate its effectiveness
The final model will be deployed as a in public spaces such as malls,
software application that can be airports, and offices. Feedback from
integrated with camera systems for these tests will be used to fine-tune
detection. The deployment strategy the model and deployment strategy.
will consider computational efficiency
and the ability to function in diverse
lighting and environmental 5.Milestones
conditions. M1: Data Collection and Preparation
(Week 1): Accumulate and prepare
the dataset for training, including
augmentation and annotation.
M2: Model Selection and W5: Princess and Parv will lead the
Development (Week 2): Conduct finalization and documentation
studies to select the most suitable phase, with support from the team as
deep learning architecture and needed.
develop the initial model.
Our objective is to implement a top-
M3: Training and Initial Testing (Week tier face mask detection system,
3): Train the model using the adhering to this outlined plan,
prepared dataset and perform initial ensuring it aligns with the project
testing to assess performance. goals and is completed within the
designated timeframe.
M4: Performance Evaluation and
Comparative Analysis (Week 4):
Evaluate the model's performance in
7.References
various conditions and compare it
with existing solutions. • Face Mask Detection ~12K
Images Dataset. (2020, May
M5: Finalization and Documentation 26). Kaggle.
(Week 5): Document the project https://www.kaggle.com/datas
including methodologies, findings, ets/ashishjangra27/face-mask-
and future improvements and also 12k-images-dataset
creating the presentation for the • S. (2022, August 17). Face Mask
same. Detection. Kaggle.
https://www.kaggle.com/code
/sayamkumar/face-mask-
6.Task Assignment detection
• Team, K. (n.d.). Keras: Deep
W1: Sumit and Parv will be Learning for humans.
responsible for data collection and https://keras.io/
preprocessing.
W2: Princess will handle feature
extraction using CNN models.
W3: Sahil will work on model
development and training.
W4: Sumit and Sahil will conduct
model evaluation and fine-tuning.