Automatic Number Plate Recognition (ANPR) System

Overview

The ANPR system detects and recognizes vehicle license plates from images or video streams. Built for Indian license plates, it handles diverse plate formats, various lighting conditions, and challenging scenarios.

Tech Stack:

• Detection: YOLOv11 (fine-tuned for license plates)
• OCR: RapidOCR (ONNX runtime, ~50-100ms per plate)
• API: FastAPI REST endpoints
• Frontend: Alpine.js + Tailwind CSS
• Database: SQLite

Requirements

• Python 3.10+
• NVIDIA GPU with CUDA 12.1 (recommended)
• 8GB RAM minimum

Quick Start

# 1. Create virtual environment
python -m venv .venv
.venv\Scripts\activate  # Windows
source .venv/bin/activate  # Linux/Mac

# 2. Install dependencies
pip install -r requirements.txt

# 3. Download YOLOv11 license plate model
python scripts/download_plate_model.py --variant s

# 4. Start the server
python src/main.py --mode api

The web UI is available at http://localhost:8000

Usage

Web Interface

1. Open http://localhost:8000
2. Upload a video or image
3. View detected plates in Results

API Endpoints

# Health check
curl http://localhost:8000/api/v1/health

# Process image
curl -X POST http://localhost:8000/api/v1/image/process \
  -F "image=@plate.jpg"

# Upload video
curl -X POST http://localhost:8000/api/v1/upload \
  -F "file=@video.mp4"

# Get detections
curl http://localhost:8000/api/v1/detections

# API documentation
open http://localhost:8000/api/v1/docs

CLI Mode

python src/main.py --mode cli

Configuration

Copy config.template.yaml to config.yaml to customize:

models:
  yolo_model: yolov11s                           # n/s/m/l/x variants
  yolo_weights: models/yolov11s-license-plate.pt
  yolo_confidence: 0.4
  yolo_fp16: true                                # GPU acceleration

api:
  host: localhost
  port: 8000

Environment variables override config file:

• ANPR_API_PORT - API port
• ANPR_MODELS_YOLO_CONFIDENCE - Detection threshold
• ANPR_MODELS_YOLO_FP16 - Enable/disable FP16

Project Structure

ANPR/
├── src/anpr/
│   ├── api/          # FastAPI endpoints
│   ├── cli/          # CLI interface
│   ├── core/         # Detection, OCR, video processing
│   ├── database/     # SQLite models and queries
│   └── config/       # Configuration management
├── static/           # Frontend (Alpine.js + Tailwind)
├── models/           # YOLO weights
├── data/             # Uploads, snapshots, database
└── scripts/          # Utilities

Development

# Run tests
python test_integration.py

# Check CUDA availability
python scripts/check_cuda.py

# Demo pipeline on sample images
python scripts/demo_pipeline.py

Performance

On NVIDIA RTX 4050 (6GB):

• YOLOv11s inference: ~10ms per frame
• OCR extraction: ~50-100ms per plate
• Full pipeline: ~100-150ms per frame
• FPS: ~90-100 (detection only)

Troubleshooting

CUDA not detected:

python -c "import torch; print(torch.cuda.is_available())"

Model not found:

python scripts/download_plate_model.py --variant s

Memory issues:

• Reduce yolo_input_size to 416
• Disable yolo_fp16 if GPU memory is limited