Research On License Plate Recognition Algorithm Based On Information Fusion

Pattern recognition is an important constituent of computer vision technology, providing effective methodology and techniques for the question of classification. License Plates Recognition technology, a heated and difficulty research issue, is a successful application of related principles and methods of pattern recognition in our daily life, continuing being improved. The results of recognition are always not satisfying using a sole classifier because of all sorts of objective factors. Based on information fusion technology, the thesis fuses the results of several classifiers or more features extracted by different methods, and get the final better recognition results.License Plates Recognition system is composed by the following parts: images collection, license plates location, character segmentation and character recognition. That how to design effective classifiers and recognize characters under the circumstances that character segmentation effect may be bad is a difficulty in license plates recognition. Analyzing of the current license plate recognition algorithm for a variety of deficiencies, The thesis proposes license plates fusion recognition algorithm based on information fusion technology. Firstly, the thesis recognizes the segmented characters using model matching method and feature matching method, getting two sets of matching scores, which come from the recognition results. Then, the thesis uses weighted matching scores fusion algorithm based on decision level and get the final matching score marked weighted fusion result. On the basis of weighted fusion result, the thesis recognizes the character using BP net and gets another set of recognition results marked NN recognition result. Based on the two sets of results, weighted fusion result and NN recognition result, the thesis designs a new fusion recognition method combining D-S theory, getting the final fusion recognition results. The thesis makes an attempt from the aspect of classifier fusion in license plate recognition.The experimental results show that the fusion algorithm of this thesis has better recognition results, with a larger increase of the recognition rate of vehicle licensing and the recognition rate of all the characters. The fusion recognition system has better effectiveness and robustness. To a certain extend, the fusion recognition algorithm can correct the false recognition results while two classifiers both have weak recognition results. In the following part, the thesis introduces the DSP developing platform, and details the framework and developing flow of embedded license plate recognition based on DM642 platform, finishing the transplantation of license plate recognition algorithm.