Adaptive Threshold Edge Detection And Machine Learning

Automatic license plate recognition technology has played an increasingly important role in the field of intelligent transportation, and is one of the important research direction in the current intelligent traffic management research. The requirement for the performance of social management system is becoming higher and traffic environment is becoming more complex, existing automatic license plate recognition technology can’t meet the needs of users, so how to recognize plate accurately and timely in complex environment is still a focus in the study of intelligent transportation researchers. This article is divided into the license plate location, license plate character segmentation and character recognition.License plate localization research how to find out the license plate and segment it from a picture. The whole image has difference in color and brightness by the influence of different time and weather. This article uses edge detection and morphological processing to locate plate, and adjust the edge weighted gradient value and binarization threshold adaptively according to the positioning effect to candidate area, and then combined with the feature of other plates to select candidate area. This approach locates more accurate.The difficulty of Character segmentation is how to eliminate the interference of license plate frame, and how to overcome the phenomenon of connections between characters. This article used the Radon transform and character width information to correct tilt, and uses character width value forecasting and validation method to determine the character stroke width, then devides the character according to this width. The Character recognition part focus on character recognition accuracy and speed, as well as the ability of recognizing imperfect license plate character. In view of the disadvantage of slow recognition method based on online sequence extreme learning machine. It improves recognition rate through training real timely, and the learning ability can be improved by the time. In view of the recognition of the incomplete characters, this paper introduces hysteresis characteristics to make network have associative memory ability, and designed hysteresis characteristic self-choose extreme learning machine algorithm. This method enhanced the network generalization ability, and improved the identification accuracy.Cui Leitao(Control Science and Engineering)...