Research On License Plate Recognition Algorithm In Intelligent Transportation System

Intelligent Transportation System (ITS) can effectively improve the trafficmanagement efficiency, transportation safety, road network capacity and transportefficiency. License Plate Recognition (LPR) technology is the crucial component of ITS.This paper does a deep research on LPR technology, which is the technology of licenseplate location, character segmentation and character recognition. We have a research onthe analysis of edge density based license plate location algorithm, connected componentanalysis based character segmentation algorithm and propose the HOG feature basedcharacter recognition algorithm.Characters in the license plate region are arranged regular and tight thus haveintensive edge, while the other parts have loose and irregular ones. The analysis of edgedensity based license plate location algorithm uses this characteristic to locate licenseplate. First, we extract the vertical edge of the image, and then analyze the density of edgeto mark the likely license plate region. Finally, do morphological processing in the markedlicense plate region and remove the false plate parts.Connected component analysis based character segmentation algorithm firstlycorrects tilted vehicle image by Hough transform line detection to calculate tilted angleand corrects image. And then we analyze connected component in the binary image tosegment characters’ region based on its characteristic.HOG feature based character recognition algorithm uses HOG feature to recognizecharacters. First, we should extract HOG feature from character’s image and then convertinto binary codes which represent the "fingerprint" information of character image. In theprocess of recognition, we should calculate "fingerprint" information from characterimage and compare Hamming distance with each templates feature in the library, find the minimum distance of template character, which is the final recognition result.In order to verify the accuracy of proposed algorithms, we run an experiment with500vehicle images. The experimental result shows that the algorithms can locate licenseplate and segment characters accurately, and have higher accuracy and recognition speedwhich has a bright application prospect.