Research And Realization Of License Plate Recognition Method Under Complex Conditions

By the end of February 2018,the number of motor vehicles in China has reached 310 million.Due to the sustained and rapid growth in the number of motor vehicles in our country,the increase of the plate type,urban motor vehicle intelligent management requirements will need to be improved,and the license plate recognition technology which has one-to-one correspondence with vehicle information needs to improve its universality.So,research on complex conditions such as excessive contrast change,the license plate tilts caused by the shooting Angle,the license plate character is blurred by long sunshine and Non-license plate information interference appears in the background,and the extraction of these license plate regions has important market value.The main content of this article is as follows:1.License plate location algorithm with interference information.This paper studies the color and edge detection based on threshold judgment combined with localization algorithm,the algorithm first extracted from the sample image with the color of the license plate color similar area,and then the threshold judgment if interference factors too much or too little candidate area to detect the edge information,and then use the morphology and the connection method of the custom form candidate connected domain,coupled with a priori information section for candidates connected domain filtering and tilt correction,finally,after the above processing,the regional HOG feature extracting and using the SVM classification method to eliminate false license plate area.And more in-depth study on the extraction of license plate area,the paper added to achieve the license plate location algorithm based on maximum stable extremal regions to extract the license plate area,first the MSER of license plate image are extracted,using the Chinese characters classifier planting seeds point and filtered to MSER regions,finally,the characters of the region merging for the full license plate area.The recognition accuracy of the two localization algorithms under complex conditions was 86.4%and 93.2%respectively.2.Character segmentation algorithm.The connected domain of license plate is divided into regions and then binarized,and the connected domain of license plate characters is segmented and extracted by combining prior information with prior information,coupled with the license plate characters behind the top two characters with the characteristic of the large spacing between five characters,the separate identification to the position of the Chinese characters,this method can split in contrast change and Chinese characters appear larger adhesion under the condition of complete segmentation of the characters.The segmentation accuracy of the improved character segmentation algorithm under complex conditions is 91.2%.3.License plate character recognition algorithm.LeNet-5 network prototype cannot be directly applied to identify Chinese characters,Numbers and letters,so in this paper,the traditional network model is improved and applied to the complex conditions of license plate character recognition,final accuracy reached 99.166%,and the whole algorithm of recognition accuracy is 91.875%,results show that the license plate localization algorithm in this paper,under the condition of complex can achieve better recognition accuracy and has good universality...