Research On The Key Technologies Of License Plate Recognition

Along with the development of economy and society, the quantitative rise of cars make a more and more urgent need for the intelligent traffic management. Plate recognition system which is a powerful weapon to achieve intelligent management has become a focus issue of the modern traffic development.License plate recognition system is an intelligent traffic management system, based on digital image processing, computer vision, pattern recognition and artificial intelligence and many other technical areas. Its main works are divided into vehicle license plate location, character segmentation and character recognition. According to the three key technologies, the main research works of the article are as follows:At the license Plate location Phase, the algorithm uses the grid-based binarization method, combined with local otsu binarization to process image. And then by locating connected component within the image, we could locate a plate precisely by aligning the location of the characters according to the standard specifications for the plates and verify by the small characters. Especially, with an improved Bernsen algorithm, binarization of the images is able to significantly reduce the effects of non-uniform illumination. Experimental results show that the method is rapid and effective, with the rate of correct location is above 98.3%.An improved image enhancement method is used to enhance the character details of the low-resolution image first at the character segmentation phase. Horizontal tilt angle can be got by the line through the center of the connected components and Vertical correction can be done by projection and pixels translation. At last, we can segment the characters using projection method after removing noise according to the standard specifications for plates.In the character recognition process, a code-recognition method based on structure and improved template matching method and the improved BP neural network. After extracting a variety of features for hiragana, we make the recognition using SVM classifier and BP network.In addition, for the address recognition, we treat the whole address as a unit, which not only greatly reduces the number of recognition categories and obviously enhances the features, but also reduce post-processing complexity. At last SVM classifier is used to classify for address. Experimental results show that recognition methods are able to achieve a good recognition effect, with the rates of recognition for number, hiragana and address are 99.9%, 98.4% and 98.8%.