Vehicle-Detection-And-Plate-Recognition

License Plate Recognition is a computer vision project that aims to detect and recognize license plates in images. This project utilizes the YOLOv3 algorithm for vehicle detection and provides multiple methods for license plate localization and character recognition.

Table of Contents

• Introduction
• Installation
• Dataset
• Pretrained Models
• Methods
• Performance
• Authors

Introduction

The License Plate Recognition project is designed to detect and recognize license plates in images of vehicles. It utilizes the YOLOv3 algorithm for vehicle detection and provides different methods for license plate localization and character recognition.

Installation

1. Download the Notebook
2. Upload Project Folder to Drive
3. Upload Notebook to Colab and Run

Dataset

The project relies on a labeled dataset of vehicle images with license plates for training the YOLOv3 algorithm. Unfortunately, due to licensing restrictions, we are unable to provide the dataset. However, you can use your own dataset or search for publicly available vehicle datasets online.

Pretrained Models

We provide pretrained models for vehicle detection and character recognition. You can download the models weight file from the following links:

• YOLOv3 Vehicle Detection Model
• YOLOv3 Car Plate Localization Model
• YOLOv3 Bike Plate Localization Model

Please place them in the Project directory. If Any Error Download weights from Project folder Here

Methods

License Plate Localization

This project provides three methods for license plate localization:

1. Custom Trained CNN: A custom convolutional neural network trained specifically for license plate localization.
2. Contour-based Localization: Utilizes contour detection techniques to locate the boundaries of license plates.
3. Haar Cascade Classifier: Uses Haar-like features and a cascade classifier to identify license plate regions.

Character Recognition

This project offers two methods for character recognition:

1. Custom CNN Model: Trains a custom convolutional neural network model for recognizing characters on license plates.
2. Transfer Learning with MobileNetV2: Adapts the pre-trained MobileNetV2 model for license plate character recognition.

Method 1 : Custom CNN Procedure

1. Defines a CNN model architecture suitable for character recognition.
2. Compiles the model with appropriate loss and optimizer.
3. Sets up data generators for training and validation datasets.
4. Trains the model on a dataset containing character images from license plates.
5. Implements custom callbacks for monitoring training progress and stopping based on criteria.

You can customize the model architecture, hyperparameters, and training process according to your specific dataset and requirements.

Method 2 : Transfer Learning with MobileNetV2

1. Training a CNN model with MobileNetV2 architecture on a dataset of license plate character images.
2. Saving the trained model and label encoder.
3. Recognizing characters on license plates from images using the trained model.

You can customize the model architecture, hyperparameters, and training process according to your specific dataset and requirements.

Performance

The performance of the license plate recognition system may vary depending on the quality of input images, lighting conditions, and other factors. In our evaluation, the system achieves an accuracy of 78% on a benchmark dataset.

Authors

• Swarup Kanade @swarupkanade
• Omkar Kanade @omkarkanade