Crowd Flow | Devpost

Real time pressure graph
Waveform Analysis
Front Page
Barriered simulation
Max agent simulation
Venue Selector

Inspiration

During the Astroworld Festival in 2021, a fatal crowd crush killed eight people during the tradegy and 2 more in the following days. Initial plans for the event contained contingencies for a wide range of scenarios, however there were no plans for crowd surge or mosh pit safety.

In 2022, a crowd surge during Halloween celebrations in Seoul resulted in 159 deaths. The police later stated that they did not have a crowd control plan in place.

These tradegies could have been avoided, if authorities and event organisers were given the necessary information to make their event as safe as possible. This is the motivation for our project Crowd Flow.

What it does

We model high attendance events such as concerts and provide organisers a predictive framework to safeguard their communities. Our project uses physics to model thousands of event attendees. Our project can analyse waveforms of music and then present this in the model in real time. During hype moments the crowd will act more erratically, while in calmer songs the crowd is more subdued and attendees will wander around more.

This gives an event organiser a key insight into how and when dangerous situations can form. Furthermore, attendee wellbeing is effectively visualised for organisers with each attendees' physical pressure colored according to how cramped they are and a graph showing in real time the average pressure being felt by attendees.

How we built it

Built entirely in Python using Streamlit for the frontend.

Challenges we ran into

Ensuring the user interface felt the same as using a web app. Streamlit usually refreshes the current page whenever any action is completed. Bypassing this was a significant challenge we faced.

Accomplishments that we're proud of

Building a fully functional predictive framework, even when we restricted ourselves in terms of user interface. The restriction on user interface allowed us to focus more on our model and the backend.