Researech Of License Plate Character Recognition Based On Machine Learning

License Plate Character recognition is an important research topic in intellectual transportation system. Vehicle license recognition, which include the following procedure:sampling, analysis, extraction, partition and recognition, is the critical technology in the environment of transportation crossroad and highway monitoring. Vehicle license recognition technology is to integrate with image process, pattern recognition and computer vision, therefore, research on vehicle license recognition technology own theoretical meaning and practical value.This paper elaborates research status of vehicle license recognition and lists some concerned basic knowledge. First, this paper shows the basic concept in machine learning and especially focus on the ensemble learning algorithm AdaBoost. Second, research for license plate location, character partition and character recognition are analyzed in detail, meantime, it gives explanation to the pre-process method for vehicle license image in our work. Finally, based on ensemble learning and the existed character recognition approach,such as weight plate match and algorithm AdaBoost, vehicle license recognition model in our paper are presented. Vehicle license recognition model utilizes pre-process method to deal with vehicle license sample in advance and then try to recognize license character by using algorithm AB-TM. To demonstrate algorithm AB-TM, algorithm template match and AdaBoost are firstly introduced in this paper. Because there exist some question and fault in classical algorithm AdaBoost, such as singleness of haar feature and performance degradation, algorithm AB-TM is presented based on fast boosting weight plate match method. Firstly, each level weak classifier build on sample weight plate, and then new classifier is created by new sample’s weight which is determined by weak classifier’s error-ratio. This procedure will continue until a satisfactory classifier is gained. Based on the theoretical analysis and experiment result, algorithm AB-TM is feasible and achieves the research aim.AB-TM owns good practical performance and can apply to different recognition domain.At last, algorithm AB-TM is partly transplanted into DM642hardware platform, which include image collection module, memory extended module and Ethernet interface module.By the experimental result of practical environment, algorithm AB-TM shows good stability and meets actual demands.