Crowd Analyser App

displays photograph taken and number of people detected in photograph

Inspiration

It was difficult to tell how many people are currently at the Comm Hall during peak hours (breakfast and dinner times). Heading to the Comm Hall when it was too crowded led to queuing up for extremely long times just to collect dinner.

Project Description

Upon entering the web app, counts the number of people at a location on-demand.

The app takes photos from a webcam.
App sends the photo to Microsoft Computer Vision to detect how many people are in that photo.
This information is sent to a web app to display the photo and the number of people in that photo.
This process automatically repeats approximately every 5 seconds.

How we built it

We utilized a Flask backend and a React front-end.
We utilized OpenCV to capture photos using a webcam;
We utilized Microsoft Azure Computer Vision to process the images and detect how many people are in the captured photograph.

Implementation Challenges

The group comprised 4 back-end developers who had minimal to no knowledge of React, which contributed to greater difficulty in developing the React web app and integrating it together with the Flask backend and Computer Vision API.

Accomplishments that we're proud of

Managed to fully integrate a working backend that can both detect the number of people in a photograph to a degree of accuracy, as well as automate the capturing of photographs with a computer webcam.
Furthermore, also learned the fundamentals of React to come up with a simple web app to display the results.

What's next for Crowd Analyser App

Can consider setting up in various locations around RH, such as the Upper Lounge and Gym.
Integrate multiple webcams in the Comm Hall to give a more accurate result.
Improve front-end React web app to automatically refresh the crowd number every 30 seconds instead of 5 seconds.
Configure API to return image URI instead of locally saving a copy of jpg image