The Improvement Research On Key Algorithms Of License Plate Recognition

License Plate Recognition(LPR), which is widely used in parking lots, exits and entrances of schools, high-ways and traffic cameras, is a key part of Intelligent Transportation System(ITS) and takes an important role in the fields of Pattern Recognition(PR) and Computer Vision(CV). Due to the ever-changing and complex working environment, LPR needs further study in spite of numbers of breakthroughs made by researchers in this field.This paper generalizes the relevant theories, algorithms, difficulties, applying scenarios of LPR and characteristics of license plate characters(LPC). Furthermore, three fundamental parts of LPR algorithms—license plate detection(LPD), segmentation of license plate characters(LPCS) and recognition of license plate characters(LPCR), have been studied and improved in the paper. The main work includes:(1)LPD:raises three methods to removes the pseudo areas detected by LPD:a) the first method based on the colors on license plate; b) the second approach on basis of the Connected Components Analysis(CCA) of LPC; c) the third one based on CCA of LPC and edge detection. The experiments result reveals that the third approach has the highest rate(99.27%) in removing pseudo areas and the detection rate of LPD is increased to 98.99%.(2)LPCS:applies a series of approaches to pre-process the license plate area and completes two methods to segment LPC:a) the first method based on Conditional Random Fields; b) second one based on Template Matching. The segmentation results shows that the second method has a better rate(94.54%).(3)LPCR:completes a method based on Support Vector Machine(SVM) and Local Binary Pattern(LBP) to recognize the LPC and raises an approach based on Template Matching to identify the error-prone characters in the first stage of recognition. The results show that Template Matching is effectively to improve the rate of LPCR to 95.40%.The above algorithms wholly meet the requirement of real-time. The LPR system in the paper developed on the platform of Visual C++ 6.0 could fulfill a task to detect and recognize license plates in a pictures and in a video real time. The system achieves the rate of 89.24% in 688 pictures and beats the target set previously.