运动模糊车牌识别关键技术研究

发布时间：2019-01-06 15:39

【摘要】：目前,车牌识别系统已广泛应用于高速公路收费管理、城市交通路口车辆违章监控、停车场管理等领域,成为智能交通系统的重要组成部分。本文针对相对运动造成的模糊车辆图像,研究其复原、定位、倾斜校正、字符分割及字符识别等关键技术,在分析近年来国内外研究的基础上,对有关算法进行了改进。首先在分析了现有的运动模糊图像复原算法的基础上,提出了一种基于点扩散函数估计和耦合梯度保真项自适应全变分算法的综合复原算法。先利用Radon变换和小波重构后的频谱图的水平投影估计出运动模糊方向和长度,再用基于耦合梯度保真项的自适应全变分算法进行复原处理。实验结果表明,该算法能较准确的估计出运动模糊参数,并取得较好的复原效果。在车牌定位阶段,由于运动模糊复原后的车辆图像清晰度与原始图像仍有一定差距,传统定位算法的准确率会受到影响,本文提出了一种先运用混沌自适应遗传算法对车牌进行粗定位,再采用投影法对车牌进行精确定位的综合算法。实验结果表明,该算法在车辆图像清晰度较差的情况下仍有较好的定位结果。在倾斜校正阶段,由于垂直倾斜对下一步字符分割的影响更大,所以对水平倾斜校正采用Radon变换法,而垂直倾斜校正则采用精度较高的旋转投影法。实验结果表明,该算法精度高、速度快。在字符分割阶段,首先运用投影法对字符进行粗分割,对于粗分割处理不了的矩形区域采用改进的连通域区域生长法进行二次分割,大大提高了分割的准确度。在字符识别阶段,采用了一种基于小波核函数的LS-SVM字符识别算法,重点分析了其中的特征提取、分类器的设计。最后通过实验比较了该算法与传统字符识别算法的性能。
[Abstract]:At present, license plate recognition system has been widely used in the areas of highway toll management, urban traffic intersection vehicle violation monitoring, parking lot management and so on. It has become an important part of intelligent transportation system. In this paper, the key technologies such as restoration, location, tilt correction, character segmentation and character recognition are studied for the blurred vehicle images caused by relative motion. Based on the analysis of recent studies at home and abroad, the related algorithms are improved. Firstly, based on the analysis of the existing motion blur image restoration algorithms, a comprehensive restoration algorithm based on point diffusion function estimation and coupled gradient fidelity adaptive total variational algorithm is proposed. Firstly, the motion blur direction and length are estimated by the horizontal projection of Radon transform and wavelet reconstruction, and then the adaptive total variation algorithm based on coupled gradient fidelity term is used for restoration. Experimental results show that the proposed algorithm can accurately estimate motion blur parameters and achieve better restoration effect. In the stage of license plate location, the accuracy of the traditional location algorithm will be affected because of the difference between the clarity of the vehicle image and the original image after the motion blur restoration. In this paper, a synthesis algorithm is proposed, which firstly uses chaotic adaptive genetic algorithm for rough location of license plate, and then uses projection method for accurate location of license plate. The experimental results show that the algorithm still has better localization results when the definition of vehicle images is poor. In the tilting correction stage, the vertical tilt has more influence on the next step character segmentation, so the horizontal tilt correction adopts the Radon transform method, and the vertical tilt correction adopts the rotation projection method with high precision. Experimental results show that the algorithm has high accuracy and high speed. In the phase of character segmentation, the projection method is used to segment the characters, and the improved connected region growth method is used to segment the rectangular region which can not be processed by rough segmentation, which greatly improves the accuracy of the segmentation. In the stage of character recognition, a LS-SVM character recognition algorithm based on wavelet kernel function is adopted, and the feature extraction and classifier design are analyzed. Finally, the performance of the algorithm and the traditional character recognition algorithm are compared by experiments.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495

【相似文献】