Eyrie - Preventing Crowd Crush Before It Happens

Best Use of Cloudflare at HackHarvard '25🏆

A project built to monitor high-density gatherings in real-time, detect dangerous crowd formations, and predict deadly crowd crush events before they occur using AI-powered drones.

Overview

Eyrie automatically streams live video from drones, detects people using YOLOv8, calculates spatial density, and generates predictive alerts. It helps authorities prevent crowd crush tragedies by providing critical minutes to respond before disaster strikes.

Features

• Multi-Source Streaming – WebRTC video from drones or cameras with real-time person detection overlays
• YOLOv8 Detection – Uses pre-trained or custom models to detect and track individuals with bounding boxes
• Spatial Analytics – Computes crowd density using Gaussian kernel algorithms with normalized coordinates
• Predictive Alerts – Machine learning algorithms identify high-risk formations before crush events occur
• Live Visualization – Interactive overlays with tracking points, risk heatmaps, and time-series analytics graphs
• Scalable Architecture – Supports multiple simultaneous drone feeds with shared video processing

Tech Stack

• TypeScript + React + Next.js 15 with d3.js for heatmap visualization
• Python FastAPI for APIs & aiortc for WebRTC streaming
• YOLOv8 (Ultralytics) for real-time person detection
• OpenCV for video processing & PyTorch for ML inference

Usage

cd frontend
pnpm install
pnpm dev

cd backend
python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt
cp config.env.example .env
python start_servers.py