CrowdSpace

CrowdSpace is a web app that allows event staff to localize potentially dangerous zones using a heatamp to analyze the crowd's density, communicate with their team and act as fast as possible.

Comment

Inspiration

After seeing how things unfolded due to crowd crushing during a music concert like Astroworld, and more recently in Itaewon, in South Korea, we were compelled to think of a solution in order to prevent such disasters from happening again. From our observations, there seemed to be a lack of infrastructure to properly perform crowd control. Our team has concluded that these catastrophic events could be prevented with the help of an app which could allow event staff or security guards quickly act when danger arises.

What it does

Using multiple person pose detection, our app tracks people's locations and maps them to a heatmap on the frontend. Using the heatmap, staff can visualize the current state of the location they are looking at. Users can switch between views, which correspond to camera views and locations. The application also allows staff to see the location of their team members and communicate with them via messaging or sending emergency pings.

How we built it

In the back end, we use tensorflow and the MoveNet Lightning model to perform multiple persone pose detection. Our Flask API processes GET requests to send an array of points corresponding to people's locations on the heatmap.

Our React front end sends GET requests to the Flask API to receive the coordinates and compute the heatmap. We also use Firestore to store the user's plans and favorites.

Challenges we ran into

The main challenge we ran into was the limitations of MoveNet Lightning. It is a fast model, but it only detects up to 6 people, which does not allow us to detect enough people in the crowd. Using another model, such as OpenPose, happened to be quite challenging and time consuming to set up due to outdated code, and training our own model was also not an option.

Accomplishments that we're proud of

We are proud of having created a large web app where most functionalities are implemented.

What we learned

Some of us learned how to use React for the first time and how to deal with a machine learning model.

What's next for CrowdSpace

We have many ideas to make our app more reliable. By using OpenPose or a trained model, we wish to achieve slightly more reliable results using computer vision. Also, we were already aware of the limitations of computer vision, which is why we think it would be more accurate to use techniques such as Wifi positinioning, bluetooth positioning or GPS positioning.

Built With

Share this project:

Updates