Computer Vision
Rakshak
Real-time crowd monitoring & stampede prevention
Dec 2025 – Jan 2026View source
- 200–500+
- People / frame
- 7+
- Risk metrics
- 30 min
- Forecast window
Overview
A real-time crowd-safety system that detects dangerous density build-up and forecasts stampede risk before it happens.
Detects 200–500+ individuals per frame with YOLOv8 + OpenCV, using CLAHE normalization and heatmap-based density estimation.
Multi-camera tracking via ByteTrack / BoT-SORT; a risk engine scores 7+ metrics (density, compression, velocity variance, direction entropy) to emit NORMAL / WARNING / CRITICAL states over real-time WebSocket alerts.
30-minute risk forecasting from temporal-spatial patterns, historical baselines, and z-score anomaly detection — validated for large public events.
Tech stack
- Python
- YOLOv8
- BoT-SORT
- OpenCV
- FastAPI
- MongoDB
- WebSocket
- JavaScript
Screens & results
