Covishield

• 

  Web Dashboard to see violations

• 

  The classifier in action - detecting no social distancing and no mask

Inspiration

While going for a morning jog, one of us realized that people are not following COVID norms and guidelines like wearing a mask and social distancing. Worse part it was virtually impossible for campuses like cal poly to track these violations.

So we decided to build a system that can catch violations based on a live video feed and display it on a web dashboard.

What it does

Using computer vision algorithms we detect whether a person is wearing a mask or whether in the case of a group social distancing is followed or not. Then, in case of a violation, the web dashboard is updated in real-time and shows the name of the violator, the violation, the time of the violation, and the location on a map.

How we built it

We used three classifiers - one for the mask, one for the person, and one for social distancing. The classifiers are written using OpenCV and TensorFlow with a python backend.

The web dashboard was built using React, javascript, and tailwind CSS. The web dashboard also shows a map which it does so using the Google Maps API.

Finally, the real-time data communication between the frontend and the backend is done using firebase Firestore.

Challenges we ran into

• issue with compiling TensorFlow on an M1 MacBook
• dealing with javascript asynchronous code
• Dealing with different timezones

Accomplishments that we're proud of

• working as a completely remote team across timezones
• Figuring out the Google Maps API
• successfully building and deploying an image classifier

What we learned

• Integrating google maps with react
• Real-time updating of the dashboard

What's next for Covishield

• Integration with campuses and offices
• Improved face and mask detection
• Setting up video streams to work with CCTV cameras
• Ability to sort and filter violations in the dashboard

Built With

• firebase
• javascript
• netlify
• opencv
• python
• react
• tailwind
• tensorflow