Recognition of Indian license plate number from live stream videos

Abstract

Automatic License Plate Recognition (ALPR) has been employed in many developed countries for traffic management, automatic speed control, tracking stolen cars and also in automatic toll systems for improving the traffic control. ALPR is a surveillance system that extracts the information from the vehicle license plate by capturing the images. Human intervention to recognize the license plates results in delays and cases of missing out certain violations. Automating the process from surveillance cameras is challenging as the system must deal with different angles, distance, resolution and illumination conditions. In this paper, we compare the accuracy of three well known approaches to detect license plates under Indian conditions. We have selected OpenALPR, k-NN and Convolutional Neural Networks (CNN) based techniques, adapted them for local conditions, compared their accuracy on both still images and live stream videos. Our analysis demonstrates that the CNN based approach is the most suitable for recognizing number plates in India.