Use this Link to watch demo of the protype: https://youtu.be/Er4poQrQ1-4

CrowdWise -- Predict. Prevent. Protect.

What Inspired This

On January 29, 2025, 30 people died and 60 were injured in a stampede at Kumbh Mela, Prayagraj.

In 2008, 147 people died at Chamunda Devi temple, Jodhpur. In 2013, 115 died at Ratangarh temple, Madhya Pradesh. In 2017, 23 died at Elphinstone Road station, Mumbai.

Every one of those incidents was preventable. Not with more police. Not with better barricades. With information knowing where the crowd was building, fifteen minutes before it became a disaster.

That is what CrowdWise does.

The Problem

Current crowd management at large events in India is entirely manual. Supervisors do headcounts. Officers communicate by radio. By the time someone on the ground notices a dangerous crowd build-up, it has already become a crisis.

There is no system that watches all zones simultaneously. There is no system that predicts what will happen next. There is no automated way to reroute a crowd before panic sets in.

The gap is not effort. It is the absence of real-time crowd intelligence.

What CrowdWise Does

CrowdWise is a real-time crowd management system built for high-density venues - temples, railway stations, stadiums, festival grounds.

It works in four stages:

Detect The system monitors crowd density across all venue zones continuously. Each zone shows people per square metre, capacity percentage, current status, and estimated wait time updated every 30 seconds.

Predict A forecasting model projects crowd density 15 minutes ahead. Alerts fire before the situation becomes critical, not after. Operators get a warning window to act that window is what saves lives.

Reroute When a zone reaches danger threshold, the A* pathfinding algorithm recalculates the safest alternate route in under two seconds. It avoids congested zones by assigning them a high traversal cost, so the path it finds is always the least crowded option available.

The operator sees a specific instruction: "Divert crowd from Zone B to Zone C — 41 spare capacity."

One click executes the protocol.

Instruct Digital signboards update automatically with crowd status, safety instructions, and estimated wait times. The prototype includes a full signboard simulator showing what the public-facing display would show in Hindi and English. This reduce panic among the public and gains their trust that situation is in control.

Features Built

Live Control Room Dashboard

A single-screen view showing all zones simultaneously. The heatmap grid updates in real time, shifting from green to yellow to red as density increases. Stat cards show total crowd, danger zone count, average density, and safe zone count. A live event log captures every alert and protocol activation with timestamps. An interactive simulation panel lets operators adjust density per zone using sliders [only for demonstration]the system under different crowd scenarios.

SmartRoute

The organiser uploads a venue floor plan before the event. The system renders it on an HTML5 canvas and allows zones to be mapped directly onto the blueprint. Once zones are defined, the A* algorithm maps all possible routes between them. As zone densities change, routes recalculate automatically. An AI predictions panel shows zone-specific risk forecasts and a recommendations panel gives the operator specific actions to take.

Digital Signboard Simulator

A dedicated page simulating what a public-facing signboard would display at each zone. It updates automatically based on the zone's crowd state and shows live wait times. Designed to connect directly to physical IoT signboards in a production deployment.

How I Built It

Stack

Next.js 14 with App Router
Tailwind CSS
HTML5 Canvas for blueprint rendering
A* pathfinding — custom JavaScript implementation
Recharts for live crowd trend visualisation
Custom simulation engine for real-time density modelling

The pathfinding logic

The A* implementation treats the venue as a grid where each cell is a zone. Danger zones are assigned a traversal cost of 100. Warning zones cost 10. Safe zones cost 1. The algorithm always finds the lowest-cost path, which in practice means it routes crowds around dangerous areas automatically. When zone states change, the path recalculates from scratch the whole process takes under two seconds.

Challenges

The real version of CrowdWise runs on live camera feeds, IoT sensors, and a computer vision model counting people in real time. None of that infrastructure exists in a hackathon prototype. The challenge was making the simulation behave like the real thing.

Crowd density does not fluctuate randomly. It builds gradually, spikes at entry rushes and event endings, and drops slowly. I studied crowd density research and stampede case studies to replicate these patterns in the simulation engine so the dashboard behaves like a real venue, not a random number generator.

In production, the AI agent would analyse an uploaded blueprint using computer vision — detecting walls, corridors, and entry points automatically, then converting the layout into a navigable graph for the pathfinding algorithm. In this prototype, that step is manual. Building toward that full automation is the immediate next step.

The logic is real. The data is simulated.

The production build closes that gap.

Deployment Roadmap

This prototype demonstrates the complete decision loop in simulation. A production system would add:

YOLOv8 and OpenCV for real person detection from CCTV feeds
TensorFlow LSTM model for time-series crowd prediction
Python and Flask backend with REST API
React Native mobile application for field officers
AWS cloud infrastructure
IoT people counter and Bluetooth tracking integration
Physical signboard connection via Raspberry Pi

Who It Is For

Control room operators need a single screen that shows everything and tells them exactly what to do. Field officers need push alerts with a location and a clear instruction. The general public needs simple signage that updates in real time. CrowdWise is designed around all three. It include event organisers who configure the system before an event, religious trusts managing temple crowds, stadium operators, and emergency services who benefit from early alert positioning.

Impact

India loses hundreds of lives every year to crowd disasters that follow predictable patterns. The density builds. The bottleneck forms. By the time anyone acts, people are already falling.

CrowdWise is built on a straightforward idea that if you can see the density building fifteen minutes early, you can stop the disaster before it starts.

The prototype works. The algorithm is real. The roadmap to production is clear.

It just needs to be built.

*A blueprint image has been provided and may be used in the prototype; however, participants are free to use their own preferred images.