anpr-license-plate-recognition

---

📖 Overview

This repository provides an Automatic Number Plate Recognition (ANPR) pipeline built with:

• YOLOv3 (OpenCV DNN module) for license plate detection.
• EasyOCR for character recognition.

The system detects license plates in images, crops them, applies preprocessing, and uses OCR to extract alphanumeric text. This is a portfolio-ready implementation, showcasing end-to-end computer vision and text recognition.

---

📂 Repository Structure

anpr-license-plate-recognition/
├─ src/                 # Source code (main pipeline, utils)
│   ├─ main.py          # Entrypoint with CLI
│   └─ util.py          # Helper functions (NMS, draw, outputs)
├─ configs/
│   └─ default.yaml     # Default configuration (paths, thresholds)
├─ models/              # YOLOv3 model assets
│   ├─ config/          # darknet-yolov3.cfg
│   ├─ weights/         # model.weights
│   └─ classes.names    # classes file
├─ data/                # Sample images (ignored in .git)
│   └─ README.md        # Instructions for placing datasets
├─ notebooks/           # (Optional) experiments, EDA
├─ docs/                # Documentation
│   └─ assets/          # Figures, diagrams
├─ build/               # Outputs (annotated images, logs)
└─ README.md            # Project documentation

---

⚙️ Getting Started

Prerequisites

• Python 3.10+
• Tesseract OCR installed and added to PATH (required by EasyOCR on some systems).
• GPU optional (OpenCV DNN runs on CPU by default).

Installation

# Clone repository
git clone https://github.com/AlbertoMarquillas/anpr-license-plate-recognition.git
cd anpr-license-plate-recognition

# (Optional) create virtual environment
python -m venv .venv
.\.venv\Scripts\Activate.ps1

# Install dependencies
pip install -r requirements.txt

# If requirements.txt not present, install manually
pip install opencv-python easyocr numpy pyyaml matplotlib

Folder setup

• Place YOLOv3 model files in models/:

  • models/config/darknet-yolov3.cfg
  • models/weights/model.weights
  • models/classes.names

• Place input images in data/.

---

▶️ Usage

Run the pipeline with default settings (YAML config):

python .\src\main.py --save --show

Specify arguments explicitly:

python .\src\main.py `
  --input-dir data `
  --model-dir models `
  --cfg config/darknet-yolov3.cfg `
  --weights weights/model.weights `
  --classes classes.names `
  --langs en `
  --save --show

Key options:

• --save → Save annotated outputs in build/outputs/.
• --show → Display annotated images interactively.
• --langs → Specify EasyOCR languages, e.g., --langs en es.

---

📊 Example Output

Detected plate text with confidence scores printed in terminal, and annotated images saved in build/outputs/.

[car1.jpg] 1234ABC (score=0.89)
[car2.png] 4567XYZ (score=0.83)

---

📁 Dataset & Models

• Datasets: not included. Place your own test images in data/.
• Models: not included. Add YOLOv3 config, weights, and classes to models/.
• See data/README.md and models/README.md for detailed instructions.

---

✨ Features

• License plate detection using YOLOv3 + OpenCV DNN.
• Non-Maximum Suppression (NMS) for cleaner bounding boxes.
• Plate preprocessing (grayscale, thresholding).
• Text recognition with EasyOCR.
• CLI with PowerShell examples.
• Configurable thresholds and model paths via YAML.
• Portfolio-ready repo structure and documentation.

---

📚 What I Learned

• Building an end-to-end ANPR system combining detection + OCR.
• Using OpenCV DNN to load and run YOLO models.
• Handling detection outputs, confidence thresholds, and NMS.
• Integrating EasyOCR for multilingual text recognition.
• Designing a portfolio-friendly project structure for recruiters.

---

🚀 Roadmap

• Add Dockerfile for reproducible environments.
• Expand support for YOLOv5/YOLOv8 detection.
• Integrate video stream input (real-time ANPR).
• Add unit tests in test/.
• Pretrained model links via GitHub Releases.

---

📝 License

This project is licensed under the MIT License.