车载云台摄像机的车牌识别系统研究

发布时间：2018-05-23 11:47

本文选题：车牌识别 + Adaboost算法　；参考：《浙江大学》2014年硕士论文

【摘要】：车牌识别是现代智能交通系统的重要组成部分,本文在本课题组研制的车载云台摄像机硬件的基础上添加车牌识别功能组成可移动式车牌识别系统,相较于传统的固定式车牌识别系统,应用更加灵活,是对目前广泛采用的固定式车牌识别系统的补充。本文主要针对车载云台摄像机可移动式的特点,从运动车辆检测、车牌定位、车牌倾斜校正及字符分割、车牌字符识别等方面展开研究,其具体工作如下： (1)介绍了车牌识别的硬件系统。主要分析了车载云台摄像机的基本功能以及搭建本系统的其他硬件设备,为实现该系统的软件算法部分奠定基础。 (2)研究分析了常用的运动目标检测方法。确定了本系统两种工作模式,即摄像机静止和摄像机运动。对于常用的工作模式即摄像机静止时,提出了虚拟线框帧差法检测运动车辆,实验证明该方法简单有效。 (3)研究分析了常用的车牌定位方法。提出了基于改进haar-like特征和Adaboost分类器的车牌定位方法。针对车牌区域的特点,设计了适合车牌的haar-like特征,通过大量的正负样本训练了Adaboost分类器,检测时运用积分图加速检测窗口的特征提取。 (4)研究分析了常用的车牌倾斜校正和字符分割技术。根据车牌字符具有丰富的垂直边缘而车牌周围很少有垂直边缘的特点,提出了基于垂直边缘特征主元分析的车牌倾斜校正方法,字符分割则采用基于车牌特征和连通域分析的车牌字符分割方法。 (5)研究分析了常用的车牌字符识别技术。本文首先提取出字符的Gabor特征和粗网格特征,再运用BP神经网络的方法,训练出车牌字符识别的分类器。 (6)根据实验结果总结分析了该系统有待改进和提高的地方,确定接下来的研究工作。
[Abstract]:License plate recognition is an important part of modern intelligent transportation system. In this paper, the mobile license plate recognition system is composed by adding the license plate recognition function to the hardware of the vehicle-mounted cloud head camera developed by our research group. Compared with the traditional fixed license plate recognition system, the application is more flexible, and it is a supplement to the widely used fixed license plate recognition system. This paper mainly aims at the mobile characteristics of the vehicle head camera, from the moving vehicle detection, license plate location, license plate tilt correction and character segmentation, license plate character recognition and other aspects of research, its specific work is as follows: The hardware system of license plate recognition is introduced. This paper mainly analyzes the basic functions of the vehicle-mounted cloud head camera and other hardware equipment of the system, which lays a foundation for the software algorithm of the system. 2) the commonly used moving target detection methods are studied and analyzed. Two working modes of the system are determined, that is, the camera is still and the camera is moving. A virtual wire-frame difference method is proposed to detect moving vehicles when the camera is still in common mode. The experimental results show that the method is simple and effective. 3) Research and analysis of common license plate location methods. A license plate location method based on improved haar-like feature and Adaboost classifier is proposed. According to the characteristics of the license plate region, the haar-like feature suitable for the license plate is designed. The Adaboost classifier is trained by a large number of positive and negative samples, and the feature extraction of the detection window is accelerated by integral image. 4) Research and analysis of common license plate tilt correction and character segmentation techniques. According to the characteristic that license plate characters have rich vertical edges and there are few vertical edges around the license plate, a method of license plate tilt correction based on vertical edge feature principal component analysis is proposed. Character segmentation uses license plate character segmentation method based on license plate feature and connected domain analysis. Research and analysis of the commonly used license plate character recognition technology. In this paper, Gabor features and coarse mesh features of characters are extracted firstly, and then BP neural network is used to train a classifier for character recognition of license plate. Based on the experimental results, the author summarizes and analyzes the areas for improvement of the system, and determines the following research work.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495;TP391.41

【参考文献】