The Research And Realization Of License Plate Recognition Algorithms

With the rapid development of economy, it proposes a higher demand to the traffic management that the vehicles of our country are more and more. It is very important to recognition the plate of the license. License plate recognition system must run all the day and night in outdoor, the plates collected are affected by the weather and the external devices, it causes great difficulties to the recognition because of fuzzy, tilt, fade and even blocking of the plate. Currently, there is no ideal license plate recognition technology on the market, the most are used in the limited scenarios, it shows that there is the further development space for license plate recognition.This paper researches the technology by three steps:License plate segmentation; License plate character segmentation(Character segmentation); Character recognition. The algorithm of plate segmentation is to find the license plate in the picture accurately. Firstly this paper describes the most license plate segmentation algorithms commonly used and its advantages and disadvantages, and then this paper proposes a plate segmentation algorithm based color space and waveform detection, the algorithm has effectively solved the problem that it is difficult to select threshold in traditional edge detection algorithm, and it has a very high rate of license plate background color recognition. In the character segmentation part, this paper summarized the advantages and disadvantages of the algorithms commonly used, then proposes a character segmentation algorithm which combines vertical projection and template matching, this algorithm do not need to find the left and right border of plate accurately and increases the precision of character segmentation. In the character recognition part, this paper compares the machine learning algorithms to template matching algorithm commonly used, then this paper proposes a plate character algorithm of template matching, which reduces the complexity of recognition algorithm design.The results of character segmentation and character recognition algorithm proposed can be fed back to the license plate segmentation module, which can be used to determine whether it is plate or not. Compared to other algorithms, the algorithm proposed by this paper increases the precision of plate segmentation (the number of plate segmented correctly/total), and reduces the error rate of plate recognition (the number of plate segmented incorrectly/total).