Saudi License Plate OCR Model

Overview

This repository contains a custom-trained OCR model for recognizing characters from Saudi vehicle license plates as part of a complete Automatic License Plate Recognition (ALPR) pipeline.

The model processes cropped license plate images and detects individual characters, which are then decoded into the final license plate text.

Model Details

Task: Optical Character Recognition (OCR)
Framework: Ultralytics YOLO (PyTorch)
Input: Cropped RGB image of a Saudi license plate
Output: Character detections with class labels, confidence scores, and bounding boxes

Intended Use

This model is intended for:

Automatic License Plate Recognition (ALPR)
Smart Parking Systems
Traffic Monitoring
Access Control Systems
Computer Vision Research
Educational Purposes

Repository Contents

ocr_best_v3.pt — Fine-tuned OCR model weights
class_names.yaml — Character class mapping used during inference

Training

The OCR model was trained on a private proprietary dataset specifically prepared for Saudi license plate character recognition.

To respect data ownership and licensing, the training dataset is not included in this repository.

Only the trained model weights are provided for inference and educational purposes.

Usage

Load the model using the Ultralytics YOLO API:

from ultralytics import YOLO

model = YOLO("ocr_best_v3.pt")
results = model("plate_image.jpg")

The predicted character detections can then be decoded into the final license plate string using the ALPR pipeline.

Source Code

The complete ALPR system—including license plate detection, preprocessing, OCR decoding, visualization, and application logic—is available in the corresponding GitHub repository.

Downloads last month: 27