RiyaTyagi/smarttraffic-ai

SmartTraffic AI 🚦

Speed + Helmet + Horn Violation Detector for Indian Roads

Built by Riya Tyagi | Research Project for Oxford MSc Application

---

🎯 What This Does

Upload any traffic video → Get an AI-annotated output video showing:

Feature	How
🚗 Vehicle Detection	YOLOv8 (car, truck, bus, motorcycle)
💨 Speed Estimation	Pixel displacement + FPS calibration
⛑️ Helmet Detection	HSV color analysis on rider head region
🔊 Horn Detection	Librosa spectral analysis (800–3500 Hz range)
📢 Noise Level (dB)	RMS energy per audio window
🚨 Violation Alerts	Real-time on-screen banner + stats

---

⚖️ Indian Law Context (Motor Vehicles Act, 1988)

• Section 194F: Honking in silence zones → ₹1,000 (1st offence), ₹2,000 (repeat)
• Pressure horns in Delhi: Fine up to ₹12,000
• Night honking (10 PM – 6 AM): Minimum ₹1,000 fine
• Silence zones: Within 100m of hospitals, courts, schools (40–50 dB limit)
• Problem: Only 0.22% of traffic challans issued for horn violations → This AI system automates enforcement!

---

📁 Project Structure

smarttraffic/
├── app.py               # Flask backend — detection pipeline
├── templates/
│   └── index.html       # Beautiful frontend UI
├── yolov8n.pt           # YOLOv8 model weights (copy here)
├── uploads/             # Temp uploaded videos
└── outputs/             # Processed output videos

---

🚀 Setup & Run

1. Install Dependencies

pip install flask ultralytics opencv-python-headless librosa soundfile

Also make sure ffmpeg is installed:

# Ubuntu/Debian
sudo apt install ffmpeg

# macOS
brew install ffmpeg

# Windows — download from https://ffmpeg.org/download.html

2. Place your YOLOv8 model

Copy yolov8n.pt into the smarttraffic/ folder.

3. Start the server

cd smarttraffic
python app.py

4. Open in browser

http://localhost:5000

---

🔬 How Horn Detection Works

Video Input
    │
    ├── Video frames ──→ YOLOv8 ──→ Vehicle tracking + Speed estimation
    │
    └── Audio track ──→ ffmpeg extract ──→ Librosa analysis
                                                │
                                    ┌───────────┴───────────┐
                                    │  Spectral analysis     │
                                    │  • RMS energy (loud?)  │
                                    │  • Centroid 800–3500Hz │
                                    │  • Low ZCR (pure tone) │
                                    └───────────┬───────────┘
                                                │
                                    Horn event detected?
                                        YES → Mark timestamp
                                        NO  → Skip

Combine: Horn timestamp → closest vehicles in frame → VIOLATION ALERT

---

📊 Output Stats

The system generates a full report including:

• Total vehicles detected
• Average & maximum speed (km/h)
• Helmet vs no-helmet count
• Number of horn/noise events
• Average noise level (dB)
• Horn event timeline with timestamps

---

🧪 Tech Stack

Layer	Technology
Backend	Python, Flask
AI Detection	Ultralytics YOLOv8
Video Processing	OpenCV
Audio Analysis	Librosa, SoundFile
Audio Extraction	FFmpeg
Frontend	HTML, CSS, Vanilla JS

---

📄 Research Paper

This project is the basis for a research paper:

"Smart Traffic Violation Detection for Indian Urban Roads: Real-Time Vehicle Speed Estimation and Aggressive Horn Noise Classification Using YOLOv8 and Deep Learning"

Target publication: IEEE Conference on Intelligent Transportation Systems

---

🎓 Academic Context

• Author: Riya Tyagi, B.Tech CSE (AI & ML), Galgotias College of Engineering
• Purpose: Research paper + Oxford MSc Application portfolio
• Problem addressed: Delhi noise pollution (90 dB peak vs WHO limit 55 dB)
• Real-world impact: Automates enforcement of Section 194F, Motor Vehicles Act 1988

---

"Laws exist. Enforcement doesn't. AI bridges the gap."