【作者】 杨大力；

【导师】 翁正新；

【作者基本信息】 上海交通大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 随着计算机和数字图像处理技术的发展,车牌自动识别系统己成为智能交通系统的重要组成部分。运动模糊图像车牌识别系统的关键技术在于数字图像预处理技术、车牌定位技术和智能字符识别技术。本文应用运动模糊方向和尺度检测算法、车牌定位技术、字符分割技术、神经网络识别技术等来解决车牌识别问题。确定车牌识别的环节与流程,根据各环节要求实现的功能,筛选方法和算法,对关键环节和算法通过实验,进行比较分析,对有关算法进行了改进或提出新的算法。首先,基于图像的方向微分原理以及基于自相关的点扩散函数尺度鉴别的原理,提出了求极值的优化算法,准确而迅速的鉴别出运动模糊参数(运动模糊方向和运动模糊距离)。其次,对基于车牌灰度分布特征的车牌定位算法作了改进,扩大了分割区域的宽度,综合其他方法如霍夫变换等对可能产生的多个候选区域进行筛选判别,使之对发生倾斜的车牌或者背景较为复杂的车牌图像也能正确分割。再次,基于单应性矩阵方法研究了图像校正技术。该方法的关键问题在于边缘检测及寻找边缘的交点。本文针对发生倾斜或仿射形变的车牌图像的校正提出了一种自动化的边缘提取算法。该方法通过对霍夫变换得到的直线进行合并、分类、排序等方法来自动获取图像边缘并计算相应的交点。对校正后的图像因为长宽比例不同而产生的问题,作了分析和相应处理。该方法保留了单应性矩阵恢复算法的方便性,同时速度更快,取得的效果也比较良好。最后,针对字符分割,提出了一种剔除车牌边框的算法,该方法能够有效的除去车牌图像的边框,利于后续的字符分割。更多还原

【Abstract】 The License Plate Recognition (LPR) system becomes an important part of Intelligent Traffic System(ITS) with the rapid development of computer science and digital image processing.The essential technology of the recognition of the motion-blurred images of vehicle’s license plate system lies in the digital image pretreatment technology, the plate location technology and the intelligent character recognition technology. The paper applies the identification of motion blur direction and the blur extent, the license location technology, the character division technology, the license character recognition technology based on BP networks to the LPR system. By considering the processes of the vehicle’s license plate identification, screening methods and algorithms according to the demands of each link function, the paper makes a comparative analysis of the key links and algorithms by experiments, and the related algorithms are improved or some new algorithms were put forward.Firstly, improve the identify approach for motion-blur parameters (motion-blur direction and motion-blur extent)from motion-blurred image based on directional derivation and autocorrelation method, put forward a optimized algorithm to search the extremum, which makes the solution fast.Secondly, approved the license plate location algorithm based on the characteristic of license plate’s gray-scale. Extend the width of the segmentation area and synthesize other methods such as hough transfer to screen the candidate regions to make it successful in handling the declining images or images with complicated background.Thirdly, Study the image calibration based on homography matrix method. The key of this method is to detect edges and their intersections. A new automatic edge detection method was put forward to the calibration of license plates. This method obtains the edges of pictures and their intersections through merging, sorting and ranging based on hough transformation automatically. For the problems result from different proportions of length and width, this paper also gave an analysis and made relative processing. Simulation result shows that this method has good performance as well as preserving the convenience of homography matrix calibration.At last, to the problem of character division, an edge rejecting algorithm was put forward. This algorithm can reject the edges of vehicle’s license plate efficiently, which is very convenient for the following treatments.更多还原