Scribd
Upload
16 views

Face Mask Detection Using Convolutional Neural Network

The document outlines a research project focused on developing a Convolutional Neural Network (CNN) for detecting face mask usage during the COVID-19 pandemic. The objective is to create a surveillance system that accurately identifies individuals wearing masks correctly or incorrectly, utilizing a labeled dataset for training. Key findings indicate high accuracy in controlled environments, but challenges remain in recognizing unknown classes and improving performance in real-time applications.

Uploaded by

m.razif
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
16 views

Face Mask Detection Using Convolutional Neural Network

The document outlines a research project focused on developing a Convolutional Neural Network (CNN) for detecting face mask usage during the COVID-19 pandemic. The objective is to create a surveillance system that accurately identifies individuals wearing masks correctly or incorrectly, utilizing a labeled dataset for training. Key findings indicate high accuracy in controlled environments, but challenges remain in recognizing unknown classes and improving performance in real-time applications.

Uploaded by

m.razif
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 16

Face Mask Detection

Using Convolutional
Neural Network
Mohd Razif Bin Shamsuddin
Muhammad Hafiz Bin Mohd Nasarudin
Shuzlina Abdul-Rahman
Visit Our Website
https://kppim.uitm.edu.my/
About Me
My name is Mohd Razif Shamsuddin. My expertise is in
Machine Learning, with extensive research experience in
Neural Networks and Computer Vision. As a Senior
Lecturer at Universiti Teknologi MARA, I specialize in
several key areas within the field of Artificial Intelligence,
including Computer Vision, Machine Learning, Fuzzy Logic,
Evolutionary Algorithms, and Big Data Analytics. My
academic and research endeavors are driven by a
passion for advancing AI technologies and their
applications across various domains.

Dr. Mohd Razif Shamsuddin


Research Interest
-ARTIFICIAL INTELLIGENCE
-COMPUTER VISION
-MACHINE LEARNING
-FUZZY LOGIC
-EVOLUTIONARY ALGORITHM
-BIG DATA ANALYTICS
Overview of Research
COVID-19 Impact:
Rapid global spread with
potentially deadly symptoms.
Airborne transmission requires
preventive measures like face
masks.
Problem Statement:
Many people wear masks
incorrectly or fail to wear them
at all.
Overview of Research
Objective:
Develop a surveillance system to identify Initial Performance

individuals wearing face masks correctly or


incorrectly. Improvement

Proposed Solution:
Utilize a Convolutional Neural Network (CNN) for
mask detection. Accessibility and
Cost
Train two CNN models with different optimizers on
a labeled dataset.
Methodology: Mainstream Adoption

Dataset consists of two classes:


a. Correct mask usage.
Market Impact
b. Incorrect mask usage.
Parameter tuning applied to improve model
accuracy.
Examples of Related Works

Facial
Authentication and Emotion Recognition Surveillance and Personalized Customer
Access Control Public Safety Experiences
Facial recognition technology is
Machine learning algorithms, used in affective computing to Machine learning models are Retail and marketing leverage
particularly convolutional analyze and interpret human applied in crowd surveillance to facial recognition to analyze
neural networks (CNNs), are emotions. This application is identify individuals in large customer demographics and
widely used in facial widely adopted in areas like gatherings or to monitor reactions. This data helps tailor
authentication systems marketing to measure specific locations for public product recommendations or
(Viso.ai, 2024; Shadman Sakib, customer responses, in gaming safety, including real-time targeted advertisements​
2024) for adaptive interactions, and in alerts for law enforcement​
mental health tools for stress
detection and emotional state
analysis​
Case Study
Methodology
This study focuses on creating a CNN model for
face recognition using a low-resolution input
image. The authors used the low-resolution
images dataset for this study rather than the
down sampled dataset. This is because the
down sampled dataset has many features for
the model to recognize compared to the low-
resolution dataset. Using a low-resolution
dataset helps the model to learn from the real-
world scenario.
CNN application for Face Mask Detection
Overview of CNN
Definition: CNNs are specialized deep learning models for analyzing
visual data, inspired by biological processes.

Key Components:
Convolutional Layer: Extracts features by applying filters.
Pooling Layer: Reduces dimensionality while retaining significant
features.
Fully Connected Layer: Integrates learned features for final
predictions.

Purpose: To classify images into "correct mask-wearing" and "incorrect


mask-wearing."
CNN application for Face Mask Detection
Layer Structure of the CNN
Input Shape: Images resized to 128×128128 \times 128128×128 pixels with 3
color channels (RGB).
Layers Overview:
a. Convolutional Layers:
256, 128, and 64 filters with 3×33 \times 33×3 kernel sizes.
Activation Function: ReLU for non-linearity.
b. Pooling Layers:
Max-pooling with 2×22 \times 22×2 stride to reduce spatial
dimensions.
c. Dropout Layer:
Prevents overfitting by randomly deactivating neurons during
training.
d. Fully Connected Layers:
Dense connections leading to the output layer with 2 neurons
(binary classification).
CNN application for Face Mask Detection
Model Optimization and Results
Training Details:
Optimizer: Adamax.
Loss Function: Sparse Categorical Cross-Entropy.
Epochs: Configured with varying values (3, 6).
Evaluation Metrics:
Accuracy: Achieved ≈99%\approx 99\%≈99% on training data.
Confusion Matrix:
High precision and recall for both classes ("mask-worn" and
"worn-incorrectly").
Real-Time Application: Deployed for live mask-wearing detection,
showcasing high accuracy in controlled environments.
Results
Results
Results
Results
CNN application for Face Mask Detection
Key Findings and Challenges in the Face Mask Detection System
Achievements:
High accuracy for "wearing mask correctly" and "wearing mask
incorrectly" classes in real-time.
Effective deployment of the model in controlled environments.
Limitations:
Struggles with unknown classes (e.g., "not wearing mask").
Performance issues with high-speed object detection causing
system shutdowns.
Poor accuracy in recognizing faces from far distances.
Future Directions:
Improve video stream input handling for better real-time predictions.
Enhance image dataset resolution to support long-range face
recognition.
Allocate additional development time to address system errors and
refine model accuracy.
Thank You
For Your Attention

Visit Our Website


https://kppim.uitm.edu.my/

You might also like