MaskOn

• 

  Our product logo

• 

  Download and Info Page

• 

  Making Live Predictions with Mask On

Inspiration

With the current coronavirus outbreak, businesses struggle to handle the economic toll of increased operational costs through sanitation and reduced business. A large concern of a business owner is the risk of their store becoming a “Hot Spot” for coronavirus cases. Dozens of studies have proven that the simplest way to reduce the risk of Covid-19 is to simply: Put your Mask On! So we set out to create a software solution that works with existing security cameras to make sure that customers were wearing their masks.

How it Works

The user experience starts with the tkinter windows application we have created. Individual frames are extracted from the live video feed from the webcam. These frames are reshaped, resized, and converted to a (256, 256, 3) byte array. The byte array is passed to the predict method of a Convolutional Neural Network. The results are then overlain with the video-feed and displayed.

How we built it

The core of our project was the Convolutional Neural Network designed in python with an open source Machine Learning tool called Tensorflow. A Convolutional Neural Network is a special class of NN with the unique ability to recognized features of data that has a structure where adjacent values can be grouped together (like an image!). In our Convolutional Layers we passed a few dozen "filters" over the image to detect different features of the image and then passed those filtered images to a Deep Neural Network. After forward propagating the network will converge to a single value ranging between 0 and 1 indicating the predicted class and the confidence in that predication.

The Data:

In order to train our CNN we needed to have a significantly large classified set of images of people wearing face masks and not wearing them. We found a large open source face mask dataset on Github. To further increase the generalizing abilities of the CNN we web-scrapped some tougher images to classify and added them to the dataset.

During training we used a technique called Image Augmentation, where we apply various transformations (zoom, skew, change saturation, etc.) are randomly applied to images in the training set; this allows for the neural network to learn image features even with significant distortion allowing it to better generalize to the variety of cameras and video capture devices.

What's next for MaskOn

We would create a version of this project that works with tf.js to allow for a web experience. We could also create a REST API to allow for other services to detect if a customer is wearing a mask; this could be used for automated door that only allows mask wearing customers in.

Built With

• machine-learning
• openv
• python
• tensorflow
• tkinter