Research On Recognition Arithmetic Of The Vehicle License Plate

With the development of the science, traffic management system has been more andmore informatization and intelligentialize. License Plate Recognition System emerges as animportant application in intelligent traffic. As we all know, license plate character recognitionsystem is the core of the LPRS, so character recognition algorithm is the focus of LPRS. Thepopular algorithm mainly includes Template Matching, BP Neural Network, Support VectorMachines and so on.These three kinds of algorithm can be used to recognize license platecharacter in specific condition,but they can not have a good affect in common conditions, thatis,they have problems in universality. So study on universality of license plate recognitionalgorithm in LPRS is very important.In this article, I analysis and research these three kinds of algorithm, and then do someimprovement on these algorithm to improv their universality. The conclusions are as follows:1. In the Template Matching algorithm, the template size is difficult to determine.Aiming to this problem, I set up some different template size to do experiment.In theexperiment, I compare the three template recognition time and accuracy, then I choose thebetter template to do experiment. This template is also used to extract feature.2. In the BP Neural Network algorithm, it wastes long time to train sample mode for itspoor speed of converging. I apply an improved BP(back propagation) algorithm. Throughchanging the output in S function to reduce the number of cycles required for convergenceand improve the network convergence speed.3. The SVM algorithm is a little complex, The kernel function and its parameter is hardto determine. I use LIBSVM tool case and set some different parameter to do experiment.Atlast, I chose the proper parameter to improve the property of the system.4. I provide a classification algorithm and design the new algorithm,which is combineTemplate Matching, BP Neural Network and SVM according to weighted voting principle.Experiments indicate this algorithm is better than one of them.5. Based on combination classifier I design and complete vehicle license platerecognition system in the VC++platform.