CrowdWatch — AI Crowd Density Analysis & Congestion Prediction System

An AI-powered crowd monitoring web application that uses YOLOv8 to detect people in real-time from camera feeds, calculates crowd density, classifies congestion levels, generates alerts, stores data, and sends email notifications to authority personnel.

Team

• C Hridya Ajay (ASI23CS071)
• Dayana Shiju (ASI23CS075)
• Gagana Parvathi (ASI23CS088)
• Submitted to: Ms Raghi R Menon

Features

• Real-time Crowd Detection: Uses YOLOv8n model to detect people from camera feeds
• Density Calculation: Calculates people per square meter in defined zones
• Congestion Classification: Low (< 5 people), Medium (≥ 5), High (≥ 10)
• Alert System: Automatic dashboard alerts and email notifications
• Bus Monitoring: Special mode for tracking bus entry/exit with line crossing detection
• Multi-zone Support: 5 simulated zones with individual crowd tracking
• Analytics Dashboard: Historical data visualization with Chart.js
• Mobile Responsive: Works on all devices with bottom navigation

Tech Stack

• Backend: Python, Flask, Flask-CORS
• AI Model: YOLOv8n (Ultralytics)
• Computer Vision: OpenCV
• Database: SQLite (crowd.db)
• Email: Gmail SMTP with App Passwords
• Frontend: HTML5, CSS3, JavaScript (Vanilla)
• Charts: Chart.js
• Icons: Font Awesome
• Fonts: Google Fonts (Poppins, Space Mono)

Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Configure Email (Optional)

Edit app.py and update these settings:

EMAIL_USER = "your_email@gmail.com"
EMAIL_PASSWORD = "your_16_char_app_password"  # Gmail App Password
EMAIL_RECIPIENTS = [
    "recipient1@gmail.com",
    "recipient2@gmail.com", 
    "recipient3@gmail.com"
]

3. Run the Application

python app.py

4. Access the System

• Main Application: http://localhost:5000
• Login: Use any credentials (prototype mode)
• Standalone Demo: Open prototype.html directly in browser (no server needed)

Pages & Features

1. Login (index.html)

• Simple authentication gateway
• Any credentials accepted for prototype

2. Dashboard (dashboard.html)

• Real-time crowd statistics
• Alert status overview
• Quick navigation to all features
• Recent alerts list

3. Live Monitor (live.html)

• Live MJPEG video stream with YOLOv8 detections
• Real-time people count and density
• System information display

4. Bus Monitor (busmonitor.html)

• Toggle between Normal and Bus modes
• Track passengers entering/exiting bus
• Occupancy percentage with color-coded progress bar
• Live passenger trend chart

5. Zone Map (map.html)

• 5-zone facility visualization
• Real-time zone status with color coding
• Animated markers for high congestion
• Zone list with detailed statistics

6. Analytics (analytics.html)

• Historical crowd data visualization
• Line chart: Last 50 readings
• Doughnut chart: Congestion level distribution
• Bar chart: 24-hour hourly averages
• Summary statistics (current, peak, average)

7. Alert Center (alerts.html)

• Alert history with acknowledgment
• Email notification status
• Test email functionality
• Unacknowledged alert tracking

8. Standalone Prototype (prototype.html)

• Complete self-contained demo
• Simulated crowd data with realistic patterns
• All features in single HTML file
• No server required

Configuration

Camera Settings

CAMERA_SOURCE = 0  # 0 = webcam, or IP camera URL
ZONE_AREA = 50.0   # square metres per zone

Detection Thresholds

THRESHOLD_MED = 5   # Medium congestion threshold
THRESHOLD_HIGH = 10 # High congestion threshold
BUS_CAPACITY = 45   # Bus capacity for occupancy tracking

Email Alerts

• Medium Cooldown: 120 seconds
• High Cooldown: 60 seconds
• Recipients: Up to 3 authority personnel
• Format: Styled HTML with crowd statistics

API Endpoints

Method	Endpoint	Description
GET	/video	MJPEG live stream with detections
GET	/api/crowd	Current crowd statistics
GET	/api/alerts/history	Complete alert history
GET	/api/history	Last 50 crowd records
GET	/api/analytics	Hourly aggregated data
GET	/api/zones	5-zone crowd data
GET	/api/set_mode/<mode>	Switch detection mode
POST	/api/acknowledge/<id>	Acknowledge alert
POST	/api/email/test	Send test email
GET	/api/status	System health check

Database Schema

crowd_history

CREATE TABLE crowd_history (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    timestamp TEXT,
    count INTEGER,
    density REAL,
    level TEXT
);

alerts

CREATE TABLE alerts (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    timestamp TEXT,
    level TEXT,
    message TEXT,
    people_count INTEGER,
    acknowledged INTEGER DEFAULT 0,
    email_sent INTEGER DEFAULT 0
);

Gmail App Password Setup

1. Enable 2-Step Verification on your Google Account
2. Go to Security → App Passwords
3. Create new app password named "CrowdWatch"
4. Copy the 16-character password
5. Update EMAIL_PASSWORD in app.py

File Structure

crowdwatch/
├── app.py                 # Flask backend with YOLOv8
├── requirements.txt       # Python dependencies
├── yolov8n.pt            # YOLOv8 model file
├── crowd.db              # SQLite database
├── index.html            # Login page
├── dashboard.html        # Main dashboard
├── live.html             # Live video monitoring
├── busmonitor.html       # Bus entry/exit tracking
├── map.html              # Zone visualization
├── analytics.html        # Data analytics
├── alerts.html           # Alert management
├── prototype.html        # Standalone demo
├── static/
│   ├── style.css         # Global styles
│   └── app.js            # Shared JavaScript
└── README.md             # This file

Browser Support

• Chrome 80+
• Firefox 75+
• Safari 13+
• Edge 80+

Security Notes

• Prototype accepts any login credentials
• Email passwords should use App Passwords, not account passwords
• Camera feed is accessible on local network only
• No user authentication in prototype mode

Troubleshooting

Camera Not Working

• Check if webcam is connected and not used by other apps
• Try changing CAMERA_SOURCE to different index or IP URL
• Ensure OpenCV can access the camera

Email Not Sending

• Verify Gmail App Password is correctly set
• Check if 2-Step Verification is enabled
• Ensure recipient emails are valid

High CPU Usage

• YOLOv8n is optimized but still CPU-intensive
• Consider reducing video resolution or frame rate
• Use GPU acceleration if available

Performance

• Detection Response: < 2 seconds
• Update Frequency: 2 seconds for UI, 1 second for video
• Database Logging: Every 30 frames (~1 second)
• Memory Usage: ~500MB (including model)

License

This project is for educational purposes. Use responsibly and comply with local privacy laws when monitoring public spaces.

---

CrowdWatch - Making public spaces safer through AI-powered crowd monitoring 🚶‍♂️📊