Machine learning for algerian car plate number recognition

Abstract

In the recent years, Automatic Number Plate Recognition (ANPR) systems have been deployed in many areas. They have many applications including parking automation, access control, monitoring and tracking vehicles. Various implementations of ANPR solutions exist, only none of them target the Algerian market. In this work we try to implement an ANPR system for Algerian vehicles. To build this system we used machine vision and machine learning algorithms. The system was divided into three steps: license plate extraction, segmentation and recognition. We used edge information and projections for locating the license plate. Then by detecting spaces in the plate we were able to segment it into digits. We trained a neural network on reading Algerian license plate numbers. This network was used to recognize the extracted digits. Testing the system on a data set of images captured from Algerian cars showed some good results. Although this system is not a finished product, it can be further improved for field deployment.