Current Systems

Commercial and government automatic license plate recognition (ALPR) systems are familiar and used in a variety of applications. Though the accuracy of typical systems is often quite high, the processing pipeline usually benefits from a number of important factors. In particular, a powerful flash and a controlled imaging scenario greatly simplify the initial plate detection stage, as background clutter can interfere significantly with many detection methods. Moreover, the core technique of many ALPR systems is a classification algorithm dependent on a large database and significant initial training overhead.

Project Goals

The aim of this project was to explore the license plate detection problem in a more relaxed set of images and without requiring the use of a classifier. Specifically, images of cars in parking lots were captured using the Motorola Droid. The plates were restricted to have less than an approximately 20 degree affine transformation due to viewing angle, and were all between 15 and 30 feet from the camera.

Team