sambett/tunisian-license-plate-recognition

End-to-end pipeline for detecting vehicles and license plates in traffic images and reading Tunisian plate text, with an interactive Streamlit app. # 🚗 Tunisian License Plate Recognition System Complete end-to-end deep learning pipeline for automatic license plate recognition from Tunisian traffic images. --- ## 📋 Table of Contents - Overview - System Architecture - Features - Performance - Installation - Quick Start - Project Structure - Datasets - Training - Evaluation - Documentation - Results - Limitations - Future Work - Contributing - License --- ## 🎯 Overview This project implements a **three-stage deep learning pipeline** for Tunisian license plate recognition: 1. **Vehicle Detection** - Detect and crop vehicles from traffic scenes 2. **Plate Detection** - Locate license plates within vehicle images 3. **OCR Recognition** - Extract text from detected plates (supports Arabic "تونس") **Key Achievement:** 89.2% character-level accuracy on Tunisian license plates with Arabic text support. --- ## 🏗️ System Architecture ``` ┌─────────────────────────────────────────────────────────────┐ │ INPUT IMAGE │ │ (Traffic Scene) │ └─────────────────────────────────────────────────────────────┘ ↓ ┌─────────────────────────────────────────────────────────────┐ │ STAGE 1: Vehicle Detection │ │ • Model: YOLOv8n (3M params) │ │ • mAP@50: 76.6% │ │ • Speed: ~10ms │ └─────────────────────────────────────────────────────────────┘ ↓ [Cropped Vehicle Images] ↓ ┌─────────────────────────────────────────────────────────────┐ │ STAGE 2: Plate Detection │ │ • Model: YOLOv8n (3M params) │ │ • mAP@50: 98.96% │ │ • Speed: ~15ms │ └─────────────────────────────────────────────────────────────┘ ↓ [Cropped Plate Images] ↓ ┌──────────── …