Stampede Alert

model predicting a calm crowd

Inspiration

We were inspired by real-world crowd disasters caused by stampedes that often happen unexpectedly in large gatherings. Our goal was to create a smart system that can detect early signs of stampede and alert authorities in real time to prevent casualties.

What it does

Our project processes live or recorded video streams divided into sectors to monitor crowd behavior. Using machine learning, it predicts the likelihood of a stampede and instantly sends alerts, helping authorities act quickly to manage the situation.

How we built it

We built this using Python with a deep learning model for video analysis, combined with a Gradio UI for easy interaction. We designed a sector-wise video processing pipeline that analyzes video feeds frame-by-frame and outputs alerts with visual feedback.

Challenges we ran into

Handling large model files on deployment platforms like Huggingface Spaces was tough because of size limits. Real-time processing and UI responsiveness required careful optimization. Also, syncing multiple video feeds and ensuring timely alerts was complex.

Accomplishments that we're proud of

Successfully integrating ML-based video processing with an interactive Gradio interface.
Designing a multi-sector monitoring system that can handle simultaneous feeds.
Overcoming deployment challenges and running the app smoothly on Huggingface Spaces.

What we learned

We deepened our understanding of video-based ML models, deployment constraints, and real-time system design. We also learned how important UI/UX is in delivering actionable alerts quickly and clearly.