🚦 Traffic Violation AI Pipeline

An intelligent, end-to-end computer vision web application to detect and analyze traffic violations.

---

📖 What it does

Traffic Violation AI is a comprehensive AI-powered application designed to automate the detection, classification, documentation, and reporting of traffic violations from surveillance footage. Designed for high accuracy, scalability, and robustness in real-world conditions, it identifies common infractions and creates actionable evidence records automatically.

⚠️ Problems Addressed

• Massive Image Volume: Thousands of images per camera daily make manual review impossible.
• Delayed Enforcement: Violations discovered hours or days later.
• Human Errors: Missed violations, fatigue-related mistakes, and inconsistent judgments.
• Difficult Conditions: Reduced accuracy due to rain, nighttime environments, and motion blur.
• Missing Plates: Vehicle identification challenges when number plates are unreadable.

🎯 Key Features & Real-world Impact

• Road Safety Impact: Reduces accident risks and improves compliance with traffic laws.
• Administrative Efficiency: Drastically lowers manpower costs and evidence management burden.
• Automated Evidence Generation: Captures violation frames, annotates bounding boxes, and securely logs timestamps/vehicle details.
• Dashboard Analytics: Aggregates statistics, visualizes violation trends over time, and provides searchable records for authorities.

---

⚙️ Advanced Image Preprocessing Pipeline

To handle extreme weather, motion blur, and low-light scenarios, we apply a robust preprocessing pipeline before running inference:

1. Normalization: Images resized to 640x640 and normalized for consistent YOLO inference.
2. Low Light & Shadow Fix: Gamma Correction and CLAHE (Contrast Limited Adaptive Histogram Equalization) applied to restore visibility.
3. Weather & Blur Reduction: Bilateral filtering reduces rain/sensor noise, followed by Unsharp Masking to correct motion blur.

---

📸 Demo & Features Deep-Dive

🏍️ Helmet Non-Compliance

How it works: Analyzes motorcycle riders to verify helmet usage. Identifies motorcycles, isolates riders, and uses a specialized model to detect unhelmeted heads. Examples:

👨‍👩‍👦 Triple Riding

How it works: Enforces capacity limits on two-wheelers by counting the number of riders overlapping with the motorcycle's bounding box. Exceeding two passengers flags a violation. Examples:

🚗 Seatbelt & Mobile Phone Detection

How it works: Scans the windshield area of detected cars to monitor driver compliance. Identifies if the driver is fastened with a seatbelt and flags mobile phone usage while driving. Example:

📸 License Plate OCR & Cropping

How it works: Once a violation is confirmed, the system extracts the license plate crop, applying preprocessing (grayscale, adaptive thresholding) before running an OCR engine (PaddleOCR/EasyOCR) to accurately extract alphanumeric text. A minimal crop is saved for evidence. Examples:

🛑 Stop-line Crossing & No-Parking Zones

• Stop-line Crossing: Automatically captures frames and generates evidence when vehicles cross designated stop-lines at signals.
• No-Parking Detection: Identifies stationary vehicles in restricted zones, tracks their parking duration, and logs illegal parking evidence automatically. Examples:

🔍 Vehicle Identification Without Number Plates

How it works: Identifying a vehicle without a visible number plate is a significant challenge. Our solution leverages visual embeddings and Facebook's similarity search algorithm (FAISS) to instantly find similar vehicles across millions of images, reducing manual investigation efforts.

---

📊 Analytics & Reporting Dashboard

The application features a comprehensive dashboard for traffic authorities:

• Violation frequency statistics.
• Heatmap-based traffic risk assessment and trends over time.
• Searchable violation records categorized by violation type.

---

📈 Evaluation & Future Scope

The models are evaluated on metrics like Accuracy, Precision, Recall, F1-Score, and mAP.

Future Developments:

• Cloud-based tamper-proof evidence management.
• Real-time city-wide deployment on edge devices.
• Automated challan (ticket) generation.
• Fine-tuning YOLO models on larger, localized, real-world challan image datasets.

---

🚀 How to get started

Prerequisites

• Python 3.10+
• Node.js 18+ & npm
• FFmpeg (Required for video frame extraction)

1. Clone the repository

git clone https://github.com/akashch1512/Traffic-Violations-Using-Computer-Vision.git
cd Traffic-Violations-Using-Computer-Vision

2. Backend Setup

cd backend
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Note: API health check available at http://localhost:8000/. Interactive docs at http://localhost:8000/docs.

3. Frontend Setup

cd frontend
npm install
npm run dev -- --port 5173

Access the web application at http://localhost:5173/.

💡 Usage Example

1. Open the frontend URL in your browser.
2. Upload a sample .mp4 traffic clip.
3. Click Analyze video to extract frames, run YOLO/OCR, and generate violation evidence.

---

🤝 Who maintains and contributes

• Maintainer: Abhay Kale & Akash Chaudhari
• Contributing: We love pull requests! Please open an issue to discuss proposed changes before submitting a PR.

❓ Where to get help

• Issues: Check the GitHub Issues tab.
• Documentation: Visit the Swagger UI at http://localhost:8000/docs while the server is running.

Built with ❤️ for safer roads.