Research Of Key Technology To License Plate Recognition

Chinese had entered the automobile society, traffic congestion was becoming more andmore serious, frequent traffic accidents were occurred in China, so the effective management ofthe vehicle would become increasingly important through the intelligent transportation system.License plate recognition technology that was one of the main technology of the intelligenttransportation system became a research hotspot. From the current development situation of thetechnology, fast and accurate license plate recognition was the focus of the study in the complexscenario.License plate recognition technology mainly included the license plate location, charactersegmentation and character recognition technology.In this paper, to the existing problem oflicense plate recognition algorithm, researching from the following three aspects:1.In the complex scenario, vehicle license plate locating accuracy rate would decline. Usingthe AdaBoost algorithm obtained better location effect at the same time, training weight ofsamples were increased and degenerated, that made the classification error of weak classifierenlarge in the later training, so the AdaBoost algorithm based on weighting adjustment of theweak classifier was proposed in this paper. By the principle of maximum response of classifiersand minimum classification error, and using genetic algorithm, the weights of weak classifierswere adjusted by the proposed algorithm, and then according to the iterated condition ofminimum loss function, the overall training error bound was controlled. The experimental resultsshowed that, the algorithm could effectively improve detection rate of the vehicle license plate,lower false positive rate.2.Photographing vehicle license plate in a large view, would cause serious perspectivedistortion, so the vehicle lincense plate perspective correction algorithm was proposed in thispaper. The algorithm detected firstly straight lines of vehicle license plate image, according tothe consistency of slope of straight lines to classification, then for the long side, the upper andlower edges were determined by the algorithm based to recent distance to horizontal center lineprinciple, for the short side group, two straight lines that they are farthest were firstly obtainedby the algorithm, through distance threshold, then they were back to their most nearly line, todetermine the left and right edges, Finally, using four vertices of the vehicle license plate,according to the perspective transformation, vehicle license plate was corrected. Theexperimental results showed that, the algorithm could better correct vehicle license plate that isserious perspective distortion. 3.Characters segmentation reliability was reduced because of characters spacing changeresulting in conglutinate characters, Therefore, a character segmentation method based on fuzzyGath-Geva clustering is proposed. The exponential distance of metric was used in the algorithm,which estimated pixels distribution of characters according to the maximum likelihood function,and pixels of characters were clustered with soft boundary using the fuzzy covariance matrices,to improve the effect that the adhesion circumstances caused by the spacing change ofcharactersc. The testing results showed that, the proposed algorithm could effectively segmentthe characters.4. For the problem that character recognition was effected by the illumination change andnoise, this paper put forward to character recognition algorithm based on principal visual words.The characters were recognized by the proposed algorithm based on principal visual words,through the match situation of it with the SIFT features. During matching, noise was reduced andmatching reliability was enhanced, making use of the main direction of matching points and thecorrespondence of the relative position in the two images. The experimental results showed that,the method could effectively improve the matching reliability and character recognition rate.