复杂场景下的中国车牌识别研究

发布时间：2018-05-05 06:11

本文选题：车牌识别 + 深度学习　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：随着人们生活节奏的不断加快,汽车的普及率越来越高,但是伴随机动车辆的大幅度增多,道路积压、城市拥堵、管理复杂等问题也变得更加突出。智能交通系统可以高效分配资源,提高道路通行能力和交通管理效率,从而缓解城市交通堵塞的压力。车牌识别技术作为智能交通系统的核心技术之一,是图像处理、计算机视觉、模式识别和机器学习等多个领域的交叉融合,广泛应用于交通监控、停车场管理、高速公路智能收费、电子警察等场景中。实际上,车牌识别技术还可以被嵌入到手持收费机、行车记录仪,甚至是手机等移动终端中,使智能交通更便利、更贴近生活,但由于采集设备不固定等原因,造成车牌在图像中出现的位置、大小随机,对车牌识别技术提出了更高的要求,因此,对复杂场景下的中国车牌识别系统的深入研究具有非常重要的现实意义。本文重点对复杂场景下中国车牌的定位、校正、分割和识别问题进行研究,并提出一些行之有效的改进措施。针对复杂场景下图像存在的光照不同、背景复杂等问题,本文提出基于卷积神经网络的车牌定位方法,并提出利用多颜色空间的车牌定位方法。基于卷积神经网络的定位方法通过不断训练学习网络实现定位的高精准度,基于多颜色空间的定位方法通过提取多种颜色信息提高车牌定位的精度和效率。针对拍摄角度、采集设备晃动等因素可能造成车牌倾斜的问题,本文采取基于最长直线的车牌校正方法,在字符分割之前校正车牌能够降低分割的难度。针对道路颠簸、背景噪声等因素可能造成字符模糊、边框粘连的问题,本文改进基于投影图像的分割方法,采用设置阈值的方式进行分割,可以提高字符分割的准确度和实用性。针对旋转、形变或模糊的车牌字符难以被准确识别的问题,本文提出基于长度特征的字符识别方法,通过提取字符的轮廓信息,可以简化车牌字符的识别过程,提高字符的识别率。
[Abstract]:With the continuous acceleration of people's life rhythm, the popularity rate of cars is getting higher and higher, but with the increasing number of motor vehicles, the problems of road backlog, urban congestion and complex management have become more prominent. Intelligent transportation system can efficiently allocate resources, improve road traffic capacity and traffic management efficiency, thus alleviating urban traffic congestion. As one of the core technologies of intelligent transportation system, the license plate recognition technology is one of the key technologies of the intelligent transportation system. It is a cross fusion of many fields, such as image processing, computer vision, pattern recognition and machine learning. It is widely used in traffic monitoring, parking management, intelligent toll collection of freeway, electric police and so on. In fact, license plate recognition technology can also be used. In the mobile terminals, such as handheld toll machines, recorder and even mobile phones, intelligent traffic is more convenient and closer to life. But because of the unfixed acquisition equipment, the location of the license plate in the image, the random size, and the higher requirements for the license plate recognition technology are put forward. Therefore, the Chinese license plate under the complex scene is given. The in-depth study of recognition system is of great practical significance. This paper focuses on the research of the location, correction, segmentation and recognition of Chinese license plate in complex scenes, and puts forward some effective improvement measures. In this paper, we propose a convolution nerve based on the problems of different illumination and complex background in the complex scene. The method of license plate location in the network, and the method of license plate location using multi color space. The positioning method based on the convolution neural network realizes the high precision of positioning through continuous training learning network. The location method based on multi color space can improve the accuracy and efficiency of the license plate location by extracting a variety of color information. This paper adopts the longest straight line license plate correction method, which can reduce the difficulty of the segmentation before the character segmentation. In view of the road bump, the background noise and other factors may cause the character blurred and the border conglutination. This paper improves the projection image based on the problem. The method of segmentation can improve the accuracy and practicability of the character segmentation. In this paper, the character recognition method based on the length feature is proposed for the problem that the characters of the license plate are difficult to be identified accurately. Improve the recognition rate of characters.

【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495;TP391.41

【参考文献】