M-Essa2/Tunisian-Vehicle-License-Plate-Detection-Challenge

AI challenge to detect and recognize Tunisian vehicle license plates. Using 900 annotated car images and 900 plate text samples, participants build models for bounding box detection and OCR. Top 5 teams on Dec 7 advance to finals, with solutions applied to traffic camera monitoring # 🚗 Tunisian Vehicle License Plate Detection & Recognition ### AI Challenge – Bounding Box Detection + OCR > Building an end-to-end AI system to detect and recognize Tunisian vehicle license plates for intelligent traffic monitoring. --- ## 🔗 Competition Link Official Zindi Competition Page: zindi.africa --- ## 📌 Overview This AI challenge focuses on developing a complete **Automatic License Plate Recognition (ALPR)** system for Tunisian vehicles. Participants are provided with: - 📸 **900 annotated car images** (bounding boxes for plates) - 🔤 **900 plate text samples** (for OCR training) The goal is to build: 1. 🎯 A robust license plate detection model (bounding box localization) 2. 🔎 An accurate OCR model to recognize plate characters The **Top 5 teams (by December 7)** advance to the finals, where solutions are evaluated for real-world deployment in traffic camera monitoring systems. --- ## 🎯 Objectives - Detect Tunisian vehicle license plates from car images - Accurately extract plate numbers using OCR - Ensure generalization across lighting, angle, and motion conditions - Develop models suitable for traffic surveillance systems --- ## 🧠 Technical Approach This project is divided into two main tasks: --- ## 1️⃣ License Plate Detection (Object Detection) We use object detection models to localize license plates. ### Possible Models - YOLOv8 - Faster R-CNN - EfficientDet - SSD Example (YOLO-based training): ```python from ultralytics import YOLO model = YOLO("yolov8n.pt") model.train(data="data.yaml", epochs=50, imgsz=640) ``` ### Evaluation Metrics - mAP (mean Average Precision) - IoU (Intersection over Union) - Precision / Recall --- ## 2️⃣ Optical Character Recognition (OCR) After detecting plates, we crop them and perform text recognition. ### OCR Approaches - CRNN (Convolutional Recurrent Neural Network) - Tesseract (baseline) - Transformer-based OCR - EasyOCR / …