Research On Key Techniques Of License Plate Recognition

As one of the very important technology in Intelligent Transportation System (ITS), License Plate Recognition (LPR) technology is the hot spot of recent years'research. Intelligent Transportation System plays an important role in transportation statistic,vehicle violation monitoring and vehicle tracking and locating. And license plate is the only identity of each vehicle, LPR technology is therefore the whole system's core technology, with practical and important research significance.There are three key techniques in license plate recognition, including license plate location, license plate character segmentation and recognition. On the basis of studying current research achievement, this paper gives deep research on license plate location and character segmentation techniques, using digital image processing technology combined with the actual situation.In license plate location module, taking into account the actual situation, this paper divides license plate location into two steps:vehicle location and license plate location. THIS method not only solves the problem that there is more than one license plate in one vehicle image, but also reduces the difficulty of locating license plate from a large resolution image. THIS paper presents an image enhancement and adaptive binarization algorithm which could eliminate the impact of reflective pavement and successfully locate vehicles. Next, we use top-hat transform, edge detection and mathematical morphology combining with license plate texture features to locate the license plate.In character segmentation module, this paper presents an algorithm based on projection and preferable characters. First, the method finds the preferable characters of a license plate by combining vertical projection with prior knowledge, then generate a sliding character template to split license characters. Experimental results prove that this algorithm has advantages of fast segmentation and preferably segmented characters.All of the algorithms in this paper are properly implemented and simulated successfully. This paper uses a license plate image library that contains more than 1,000 pictures as a data source and it proves the algorithm has a high speed and is relatively robust.