Skip to content

README.md

Latest commit

 

History

History
88 lines (49 loc) · 2.77 KB

README.md

File metadata and controls

88 lines (49 loc) · 2.77 KB
Raw

Addis-Ababa-Minibus-Taxis-License-Plate-Recognition-Technology

🚖 Addis Ababa Minibus Taxi License Plate Recognition

📘 Overview

This project presents a hybrid deep learning framework for automatic license plate recognition (ALPR), specifically tailored for Addis Ababa minibus taxis. The system addresses challenges caused by multilingual scripts (Amharic & Latin), diverse plate layouts, and vehicle-type identifiers, which make existing ALPR systems unreliable in Ethiopia.

🎯 Objectives

Automate vehicle identification and license plate recognition for public transport vehicles.

Reduce manual inefficiencies in route monitoring and traffic enforcement.

Develop a context-aware, scalable, and accurate ALPR solution for low-resource urban environments.

🧠 System Architecture

The framework consists of three main modules:

  1. Vehicle and Plate Detection

Model: YOLOv11-nano (You Only Look Once, version 11)

Function: Detects minibuses and localizes license plates.

Performance: Precision = 0.941, Recall = 0.955

  1. Character Segmentation

Method: OpenCV-based adaptive preprocessing with dual-projection algorithm

Function: Isolates characters from complex license plate layouts (single, double, and vertical).

  1. Character Recognition

Model: Modified LeNet-5 CNN

Function: Classifies 13 character classes (Amharic letters, Latin alphanumerics, and vehicle identifiers).

Accuracy: 99% on test data

📊 Results

Dataset: 109 real-world minibus taxi license plate images

Correctly recognized: 97 plates (88.9% overall accuracy)

Robust performance under lighting variations, occlusions, and layout diversity

💡 Key Contributions

A hybrid ALPR pipeline optimized for multilingual and low-resource contexts.

Custom segmentation algorithm for complex Amharic-Latin mixed scripts and layout variations.

Demonstrated feasibility of intelligent transportation systems in Ethiopia.

⚙️ Technologies Used

Python, PyTorch, OpenCV, NumPy, Matplotlib

YOLOv11-nano for detection

LeNet-5 CNN for recognition

Custom preprocessing pipeline

📁 Dataset Notice

⚠️ Note: The dataset used in this project contains real license plates from public minibus taxis. Due to privacy and legal restrictions, the dataset is not included in this repository. A small sample or synthetic dataset may be shared upon request for academic use only.

📈 Future Work

Expand dataset to other Ethiopian regions. Improve execution-time performance for faster and more efficient inference. Integrate real-time detection for live traffic video streams.

👨‍💻 Author Eyuel Semeon Erchmo Graduate degree Thesis, Huazhong University of Science and Technolog, may 2025. Contact: eyuel.semeon@gmail.com

🙏 Acknowledgment

my respected supervisor Associate Professor Feng He