Mask Monitor

Home Page
Upload Page
Demo image
Demo image

Team 13

Jayden Chiu - Jayden#3838
Marcus Chok - Ace no ™#8264
Steven Yau - stevenn#8855
Brittany Wong - Brittany#8474

Inspiration

We wanted to learn something related to ML/AI as we all thought it was an interesting topic we can learn for QHacks and we also wanted to tackle an issue related to the current pandemic, so we decided to make a website for business monitors to monitor if customers are wearing masks properly. We decided to use Tensorflow and OpenCV along with Flask to tackle these issues and create a website. Now, business owners can track who's wearing a mask properly depending on where they set up their camera and monitor it from a single website, making it much easier monitor potential exposure to COVID-19 in their business!

What it does

Detects if user is wearing a mask, wearing a mask incorrectly, or not wearing a mask at all from webcam or uploaded video
Displays live webcam footage on home page
Allows users to upload their own videos and download them with detected masks and violations

How we built it

Datasets

MaskedFace-Net, AI-generated face mask dataset
RWMFD, real-world face mask dataset
WIDERFACE Dataset, face dataset

Models

Finetuned model based off of MobileNetV2 for detecting mask status of face
OpenCV DNN sample for face detection model

Technologies

Google Colab GPU runtime used to train models

Stack

Flask for website backend routing
OpenCV used for video processing
Tensorflow and Keras for training models
Bootstrap CSS for UI

Challenges we ran into

Model inaccuracies depending on viewing angle, lighting, and mask
Depending on epochs, initial learning rate, etc., model would overfit to data
Choosing the best gradient descent optimization algorithm for training
Model would sometimes break because of improper frame loading
Google Colab would sometimes take excruciatingly long to load images from datasets
Bootstrap CSS would not create UI as we wanted

What we learned and our accomplishments

Learning Tensorflow and Flask for the project
Learning about how ML and neural networks work
Learning how to use Google Colab to train your own ML models
Finding proper face mask datasets to train the model with
Running OpenCV locally compared to displaying it on a website using Flask
Creating a responsive UI using Bootstrap CSS
Training a model with a high enough accuracy (~93% accuracy) that we can deploy

What's next for Mask Monitor

Create authentication system which can email users if a mask violation is detected
Allow users to upload webcam footage to the cloud so they can review the videos later
Find larger datasets to increase the accuracy of our model
Expand our app to tackle more COVID-19 related issues such as social distancing or self-isolation
Deploy the app so owners can use it from anywhere