Nepali Number Plate Character Recognition Project

This project is developed as part of the Advanced Data Science elective course offered by Samsung. The goal of this project is to recognize Nepali number plates characters from vehicle images by utilizing deep learning techniques, specifically YOLO and CNN models.

Contributors:

• Satyasa Khadka
• Sandhya Baral
• Sudip Tiwari

---

Overview

The system is designed to perform the following tasks:

1. Number Plate Detection:

   • A pretrained YOLO model is used to detect the number plate from an image of a vehicle.

2. Character Detection in Number Plate:

   • A second YOLO model, trained on the Inspiring Lab dataset, detects individual characters in the number plate. This dataset contains images of Nepali license plates, making the model well-suited for this task.

3. Character Recognition:

   • Detected characters are cropped from the number plate image and fed into a custom-built CNN model.
   • This CNN model, trained on the Inspiring Lab dataset (which includes characters frequently used in Nepali number plates), identifies the characters.

---

Key Features

• Pretrained YOLO Model: Efficiently detects number plates from vehicle images.
• Custom YOLO Model: Specifically trained to detect characters on Nepali license plates.
• Custom CNN Model: Designed for character recognition using data tailored to the Nepali context.
• Dataset: Utilizes the Inspiring Lab dataset, ensuring high accuracy for the unique structure of Nepali license plates.

---

Methodology

1. Number Plate Detection:

   • The pretrained YOLO model takes an image of a vehicle and identifies the region containing the number plate.

2. Character Detection:

   • The custom YOLO model identifies individual characters in the detected number plate region.
   • Detected characters are cropped for further processing.

3. Character Recognition:

   • Cropped characters are passed to the custom-built CNN model, which outputs the recognized character.
   • The CNN model is trained on a dataset of characters commonly used in Nepali number plates to ensure high recognition accuracy.

---

Technologies Used

• YOLO (You Only Look Once):

  • Pretrained model for number plate detection.
  • Custom-trained model for character detection.

• Convolutional Neural Network (CNN):

  • Custom model for character recognition.

• Datasets:

  • YOLO Training Dataset: Nepali Vehicle Number Plate Dataset
  • CNN Training Dataset: Nepali Number Plate Characters Dataset

---

Applications

• Automated vehicle registration systems.
• Traffic monitoring and management.
• Smart parking systems.

---

This project demonstrates the effective application of deep learning models to solve a real-world problem, leveraging pretrained and custom-trained networks for detection and recognition tasks.

Demo Video

🎥 Watch the demo video of our project:

Untitled.mov

Example of Embossed Plate detection