License Plate Localization of Moving Vehicles in Complex Scene

Abstract

In this paper, a license plate detection system is proposed to localize and extract license plates of moving vehicles in complex scene. The system works on compressed low resolution video streams obtained from real working environments. The illumination conditions are varied which typically occurs in outdoor environment. The vehicles are moving and the background scene is quite complex. The method consists of three main steps. First, input images from video are passed through Sobel edge detector with automatic variable thresholding to cope with varied illuminations. Then, a predefined size moving window is passed through the image by skipping with a certain number of pixels. For each location of moving window, the authors consider the distribution of edge pixels to determine whether it is likely to be a license plate. The license plate likelihood scores of each window are then ranked and analyzed to suppress some false detection. The experimental results showed our method achieves up to 99% detection rate on compressed video streams containing 317 moving vehicles in varied illumination conditions from morning to night time