Research Of The Video Based Embedded Vehicle License Plate Recognition System

With the burden of the city transportation system become higher and higher, people pay more attention to the intelligent transportation system. The vehicle license plate recognition system is an important part of the intelligent transportation system and it is applied widely nowadays. In this paper, we do some research on the video-based embedded vehicle license plate recognition system, which include hardware structure, software implementation, and we propose algorithms on background reconstruction in vehicle detection and license plate images enhancement in license plate recognition. Research works in this paper are summarized as follow:Firstly, we analyze the common vehicle license plate recognition system implementation, and propose to merge vehicle detection and license plate recognition in one system, so we can make the hardware system simple and reduce the cost.Secondly, we develop the background reconstruction algorithm. We find the problems in exist algorithms when scene intensity changes rapidly, and improve them. The new algorithm divide the input image into blocks overlapped with each other. Each block is regard as a multiple Gauss model. We use dynamic threshold on the mean value and static threshold on the variance for each block to form several clusters. Then, gain compensation and linear blending are applied to clusters which are candidates for background to eliminate the edges between blocks. Experimental results show that the method proposed has a good effect in different conditions.Thirdly, we use the advantage that multiple license plate images can be got from video sequence to propose the super-resolution method on the license plate images. We use regularized super-resolution reconstruction approach, and anisotropic diffusion model as a restriction. And alternately registration and reconstruction iteration method is adopted. After that, model based character recognition is applied on the reconstructed images and original images. Experimental results show that super-resolution enhancement has a good effect on the recognition.Finally, we give the hardware and software architecture of the whole system.