-
-
Guardian Vision:AI-powered crowd intelligence that detects overcrowding early and helps authorities prevent emergencies before they escalate
Inspiration Crowd disasters are so common because often people don't see a danger until it's too late. What we wanted to create was a system that would add another layer of safety to normal CCTV in crowd-heavy environments such as stations, stadiums, campuses, and public events: an early-warning system.
What it does Scar Face is able to identify people in camera images, monitor how people move within the crowd, segment the zone, determine the density of people in the zone, and raise alarms to the authorities if an overcrowding situation or surge risk is detected. It also offers a live dashboard to report incidents, respond to emergencies, and monitor safety.
How we built it Frontend Built with React, Vite, Tailwind CSS, Framer motion and Recharts. It is developed with node, express, mongodb and mongoose. We employed Python, OpenCV and person detection with centroid tracking using YOLO for AI detection.
Challenges we ran into The main challenges faced were minimizing double counting, mapping people correctly to zones, feeding real-time AI data to the dashboard, dealing with the simulation data from the back end, and getting the system to feel as though it was a real product, but not demo-ready.
Accomplishments that we're proud of We developed a fully featured AI safety platform that detects live, has dashboard analytics, incident workflows, generates alerts, has emergency response capabilities, documents, supports Docker and is ready to run on AWS. We are proud of the fact that it is addressing a legitimate public safety issue.
What we learned We learned how to link computer vision into full–stack systems, how the crowd density can be turned into an actionable alert, how to plan a project for production readiness, and the importance of clear dashboards in emergency decision-making.
So what happens when Scar Face's next generation is born? We will look to enhance the accuracy of the models, expand camera support, implement live model updates via WebSocket, deploy to AWS, add better authentication features, integrate with predictive crowd-risk analysis, and integrate with actual emergency response systems.
Log in or sign up for Devpost to join the conversation.