Research Of License Plate Recognition System Based On Machine Learning

As the number of vehicles is growing fast, the amount of the license plate data generated every day is very huge. Under the background, we should develop intelligent transport systems to assist traffic monitoring and management. Automatic license plate recognition (ALPR) is the most basic step of intelligent transportation system. This paper realized ALPR system.License plate zone recognition is one of the most important steps in the entire license plate recognition. The error identify for license plate zone usually result in license plate recognition inaccurate. Therefore, this paper using color and edge features to detect license plate, combines the advantages of color and edge and got both lower time and space complexity, while improving the recognition performance. After extraction of the candidate set of plates, we use Bag of Wordsmodel to recognize the true license plate. Firstly SIFT feature is extracted from each plate candidate. Since the number of feature points is different, we need to use k-means clustering algorithm features to form visual dictionary. Then the SIFT feature is mapping to the visual dictionary to get a histogram, and finally we use SVM on features identification and classification.License plate character recognition is the final license plate recognition step. In order to enhance the discrimination of identification features, we use both projection histogram and histograms of oriented gradients for license plate character recognition. Histogram of gradient feature is inspired by the HOG features, and achieved very good results in terms of detection and identification of local characteristics of the image. Experimental results show that the character recognition method used in this paper has very good recognition rate, achieving a correctness of 91.2% among1000 test pictures.