Automatic localization, segmentation and classification of Brazilian and MERCOSUL vehicle plates
MERCOSUL plates, plate localization, plate segmentation, classification of plate characters
The large expanse seen in the Brazilian vehicles' fleet for the last decade has impacted in a decisive manner the methods used to handle situations that were mannualy controled before. Tasks such as the monitoring and traffic control, stolen car detection and the access control to restricted areas, are usually handled with the aid of automatic systems for license plate recognition (ALPR - Automatic License Plate Recognition). This technology is used for identifying vehicles on images and videos. The ALPR systems are composed of three sequential tasks, the plate localization, followed by the segmentation of the characters and the recognition of those characters. There are many methods that were developed for each of these tasks, using image processing and machine learning techniques. The Brazilian vehicle licensing system is working on a gradual transition to the MERCOSUL standard, and both types of plates will cohexist for some years to come. In this work, we propose to search for a good combination of techniques from image processing and machine learning, as well as convolutional neural networks to build an efficient system for performing the ALPR both on the current plates as well as the new MERCOSUL plates.