Detect and Recognize Vehicle’s License Plate with Machine Learning and Python — Part 1: Detection Li

Introduction

If 20 years ago, building a robust Automatic License Plate Recognition (ALPR) system could be considered as a Master or PhD level in the field of Computer Vision. Nowadays, the rise of Machine Learning (ML), especially with the formation of Neural Network and open-source ML libraries (Keras, Pytorch, MXNet, etc), has given us significantly accessibility to duplicate and deploy these state-of-the-art programs. If you type “Automatic License Plate Recognition with Machine Learning” in Google search bar, there might be more than 10 millions related results returned. However, although it is easy to clone and run an open-source Github repository, you might wonder what happens in the system and how can you reverse engineer it to make your own application.

For this reason, this tutorial is created to give you step-by-step instructions on how to deploy a typical ALPR system using pre-trained model Wpod-Net, Computer Vision techniques with OpenCV and Character Recognition with Neural Network.

The road map of this tutorial are divided into three parts as following:

• Part 1: Implementing a pre-trained model named Wpod-Net to detect and extract License Plates of vehicle images from 10 different countries (Germany, Vietnam, Japan, Thailand, Saudi, Russia, Korea, USA, India, China).

Outcome of Part 1

• Part 2: Plate character segmentation with OpenCV

Outcome of Part 2

• Part 3: Train a Neural Network to predict segmented characters obtained from Part 2

O