Bubble Hubble

Landing Page
Main Page

Inspiration

Given the current pandemic situation with COVID-19, businesses have been hit particularly hard. As businesses have started to reopen, taking the necessary precautionary measures to ensure the health and safety of customers is crucial. Thus, what inspired us to create Bubble Hubble was to provide a service that allows businesses to safely transition to reopening their stores.

What it does

Bubble Hubble provides users with an easy to navigate web application that allows businesses to easily connect their CCTV to our services, which will be capable of detecting whether or not customers are wearing masks and partaking in proper social distancing. Through a login system, customers can sign into their account and view footage from their business’s CCTVs, along with additional analytics that directly overlay on top of the retrieved video and graphically represented data. Through our analytics, we can give business owners insight into how well their precautions are working and alert them if their customers are at significant risk.

How we built it

First, we designed the platform together on Figma. BubbleHubble was then built using HTML, CSS and Javascript on the front-end with a OpenCV + DarkNet + Flask powered Python backend.

We perform body proximity recognition based on the YOLOv3 model where we apply additional calculations on top of the set of probable body locations. It works by performing a perspective transform to scale the image accordingly so as to accommodate for the difference in absolute pixels per meter in real life between close and far objects.

The flow of the backend is as such: ingest mjpeg video stream, perform processing as we described above, then serve the image as a mjpeg output stream for the frontend to display.

Challenges we ran into

Since the fundamental premise of our project (using computer vision for contact tracing) has been done before, we were challenged to think of unique applications of the technology. This led us to think more critically about our user interface design and feature set, which differentiate our project from previous ones. We also strived to improve the precision and accuracy of the model regardless of the video conditions which was difficult in the limited time that we had.

What's next for Bubble Hubble

Some of the next steps for Bubble Hubble include improving our computer vision algorithm to better handle a variety of lighting conditions. We also hope to add more functionality to our computer vision model to provide surface contact detection so that our clients can recognize high traffic surfaces and take the appropriate precautions.