监控视频中模糊车牌图像识别关键技术研究

发布时间：2018-03-01 15:08

本文关键词： 模糊车牌识别图像模糊度评价模糊字符识别字典学习稀疏表示　出处：《南京邮电大学》2017年硕士论文　论文类型：学位论文

【摘要】：作为智能交通系统的重要组成部分之一,车牌识别在人们的日常生活中发挥着不可替代的作用。但是由于监控摄像机自身存在缺陷、恶劣天气的影响以及车辆与摄像机之间的相对运动等因素,部分采集到的车牌图像通常会变得模糊,因此就有必要建立能够识别模糊车牌图像的应用系统。本文通过对模糊车牌图像的特征进行详细分析,以及对车牌识别、图像模糊度评价、模糊图像识别等技术进行深入研究,实现了对模糊车牌识别系统的建立。本文的主要工作及研究成果包括以下三个方面:1、总结了常见的车牌图像定位、字符分割以及预处理方法,并结合需要识别的模糊车牌图像的特点,提出了适用于模糊车牌图像的定位、字符分割以及预处理方法,然后基于以上方法创建了模糊字符图像库。2、实现了对模糊车牌图像模糊度的评价。模糊度评价的目的是将严重模糊的车牌图像以及比较模糊的车牌图像区分开,从而只针对比较模糊的车牌图像设计识别模型。首先,结合人眼对模糊车牌后五位字符的识别能力建立了一个模糊度主观评价模型,并基于该模型创建了一个带有模糊度等级标签的模糊车牌图像数据库;其次,在所创建的模糊车牌图像数据库的基础上,利用投影字典对学习模型以及逻辑回归模型建立了一个模糊度客观评价模型。实验结果表明,所建立的客观评价模型取得的评价结果与主观评价模型取得效果具有较好的一致性。3、将稀疏表示分类模型以及盲去模糊模型紧密联合起来,建立了一个模糊车牌字符识别模型。所建立的模型将图像去模糊以及图像识别耦合在一个能量方程中,使得去模糊以及识别相互影响、相互促进,最终同时实现了对模糊字符图像的恢复和识别。实验结果表明,所建立的模型具有良好的去模糊以及识别效果。
[Abstract]:As an important part of intelligent transportation system, license plate recognition plays an irreplaceable role in people's daily life. Because of the bad weather and the relative motion between the vehicle and the camera, some of the license plate images are usually blurred. Therefore, it is necessary to establish an application system that can recognize the fuzzy license plate image. Through the detailed analysis of the characteristics of the fuzzy license plate image, and the in-depth research on the license plate recognition, image ambiguity evaluation, fuzzy image recognition, etc. The main work and research results of this paper include the following three aspects: 1. The common license plate image location, character segmentation and preprocessing methods are summarized. Combined with the characteristics of fuzzy license plate image which needs to be recognized, the method of location, character segmentation and preprocessing for fuzzy license plate image is proposed. Then, based on the above methods, the fuzzy character image database .2is established to evaluate the ambiguity of the fuzzy license plate image. The purpose of the ambiguity evaluation is to distinguish the seriously blurred license plate image from the relatively fuzzy license plate image. So only the fuzzy license plate image recognition model is designed. Firstly, a subjective evaluation model of fuzzy degree is established by combining the recognition ability of the human eye to the five characters behind the fuzzy license plate. Based on the model, a fuzzy license plate image database with fuzzy grade label is created. Secondly, the fuzzy license plate image database is created based on the fuzzy license plate image database. An objective evaluation model of fuzzy degree is established for learning model and logical regression model by using projection dictionary. The experimental results show that, The evaluation results obtained by the established objective evaluation model are in good agreement with the results obtained by the subjective evaluation model. The sparse representation classification model and the blind de-fuzzy model are closely combined. In this paper, a fuzzy license plate character recognition model is established, which combines image de-blurring and image recognition into an energy equation, which makes de-blurring and recognition interact and promote each other. Finally, the restoration and recognition of fuzzy character images are realized at the same time. The experimental results show that the proposed model has a good effect of de-blurring and recognition.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】