ML Video Annotations


This web app uses a machine learnining model to classify images into categories of hand and facial gestures. It then paints a specified image onto the screen for each category.

Using OBS Studio's virtual camera, you can use the browser window as your video feed in video conferencing applications and have annotations pop up in your meetings.

I used Google's Teachable Machine to train an image recognition model of me making various gestures and facial expressions and label them appropriately. Teachable Machine then allows you to upload this model and then you just plug in the URL in your code.

The actual machine learning classification is done using a library called ml5.js. It accepts the model as a parameter in its image classifier and then allows for the main functionality of classifying the video feed being shown. It then sets a label variable which is then used to draw the relevant image.

This project uses p5.js to create a canvas in the browser that displays the video feed. You can then overlay images/GIFs/videos on top of your video feed when a gesture/facial expression is detected.