基于视频技术的汽车车型识别系统研究

发布时间：2018-07-12 09:02

本文选题：智能交通 + 高斯模型　；参考：《长沙理工大学》2015年硕士论文

【摘要】：随着国家实力的快速提高,城市化持续推进,人民可支配收入的持续上涨,2014年底中国机动车保有量达到2.64亿,跃居全球第二位。在方便资源配置、人们出行的同时,也带来了日益严重的交通问题。道路拥堵、管理混乱、交通环境恶化的状况越来越引起社会和政府的重视,迫切寻求一种更为快速、高效、实用的道路管理方式,智能交通系统(ITS,Intelligent Transportation System)应运而生并取得飞速发展。汽车车型识别系统是ITS的重要组成部分,为交通管理提供了大量的数据支撑。相对于传统的配备射频装置、利用电磁感应等方法,论文采用的基于视频的图像处理技术不仅安装方便,更新简单,而且可以获得更加丰富的道路信息。通过对图像处理检测技术的研究,论文设计出一套自动识别汽车车型的实时判别系统。论文主要工作有以下三部分:(1)分割目标车辆:针对背景法提取目标模糊,帧差法容易出现空洞、改变形状,光流法计算复杂等情况,本文提出优化的高斯模型检测算法,构建基于时域和空域的背景模型,并对传统迭代分割法进行改进,选取合适阈值消除阴影,避免出现获得的目标区域大于车辆本身的情况。(2)针对单一特征难以完善表述车辆信息并加以区分的问题,本文选取纹理图像的熵、边缘密度、面积,长宽比特征来描述车辆图像,作为车辆图像的分类特征。(3)介绍了模式识别和支持向量机,提出优化的二叉树向量机分类方法,并用其对样本进行训练,确定可分性测度后,优先判别出容易区分的类别,并根据交叉验证法选择合适的和核函数与惩罚参数,最后对本方法进行仿真,结果正确率令人满意,适合用于汽车车型识别。
[Abstract]:With the rapid development of national power and urbanization, people's disposable income continues to rise. China's motor vehicle ownership reached 264 million at the end of 2014, ranking second in the world. In the convenient allocation of resources, people travel, but also brought more and more serious traffic problems. The situation of road congestion, chaotic management and deterioration of traffic environment has attracted more and more attention of the society and the government. It is urgent to seek a more rapid, efficient and practical way of road management. Intelligent Transportation system (ITS) emerged as the times require and made rapid development. Vehicle recognition system is an important part of its, which provides a lot of data support for traffic management. Compared with the traditional radio-frequency equipment and electromagnetic induction, the video-based image processing technology used in this paper is not only easy to install, but also easy to update, and more abundant road information can be obtained. Based on the research of image processing and detection technology, this paper designs a real-time discriminant system for auto-vehicle recognition. The main work of this paper is as follows: (1) Segmentation of target vehicle: aiming at the background method to extract target ambiguity, frame difference method is easy to appear holes, change shape, optical flow method is complex to calculate and so on, this paper proposes an optimized Gao Si model detection algorithm. The background model based on time domain and spatial domain is constructed, and the traditional iterative segmentation method is improved, and the appropriate threshold is selected to eliminate the shadow. To avoid the situation that the target area obtained is larger than the vehicle itself. (2) aiming at the problem that it is difficult to describe and distinguish the vehicle information with a single feature, the entropy, edge density and area of the texture image are selected in this paper. The aspect ratio feature is used to describe the vehicle image as the classification feature of the vehicle image. (3) the pattern recognition and support vector machine are introduced, and an optimized binary tree vector machine classification method is proposed. The suitable sum kernel function and penalty parameters are selected according to the cross validation method. Finally, the simulation results show that the accuracy of the method is satisfactory and suitable for vehicle type recognition.
【学位授予单位】：长沙理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495;TP391.41

【相似文献】