基于稀疏编码的车型识别方法研究

发布时间：2019-06-30 21:09

【摘要】：城市交通监控体系,为智能交通系统的主要分支之一。城市交通中,交通监管在智能化发展上发挥着核心效用,亦于交通领域以及模式识别上,皆发挥着举足轻重的作用。随着社会经济的发展,及大众生活水平提升,机动车辆呈现出逐年增加趋势,私人拥有机动车的增长,更为迅猛,由此诱发了系列交通问题。交通违章及事故增多,给城市交通及人身安全带来隐患。故此,对交通管理提出更高要求,因无法实现对城市内的所有路口与路段,实施全方位全天候的“人力监控”,要解决这问题,就必须运用“机器视觉及图像处理”科技技术,设立城市智能化交通监管体系,对过往的车辆实施车型检测及识别。本文对基于车辆图像的车型分类问题展开研究,通过多视角车辆分类器和稀疏编码,设计一个有效的车型识别方法,可对摄像机采集到的视频与图像内的车辆,实施定位与车型分类,具体如下:1)边界定位。于计算机上,分析及处理所采集到的图像,获取车辆的轮廓信息,即系车辆的长、宽、高等数据。2)采用HOG特征和SVM,构建多视角分类器,对待分类的车辆图像进行视角估计。3)对于确定视角的车辆图像,通过不同视角的数据,采用稀疏编码方式,提取有效的车型特征;每个视角下,单独进行稀疏编码。4)将每种视角下稀疏编码以及视角估计的结果,进行合并,设计最终的车型特征,采用SVM对车型进行分类,判断出车型。相比与传统的方法,本文的改进之处在于将稀疏表示引入的车型特征的表示中,通过将单一尺度的HOG特征扩展到多尺度的模式下。研究结果显示,稀疏编码方法分类的正确率高于传统词包方法,采取多视角识别手段,可提升识别正确率,经与多级分类器相结合手段,对车型图像实施多级划分,亦有效提升了分类正确率。
[Abstract]:Urban traffic monitoring system is one of the main branches of intelligent transportation system. In urban traffic, traffic supervision plays a core role in the development of intelligence, and also plays an important role in the field of traffic and pattern recognition. With the development of social economy and the improvement of public living standards, motor vehicles show an increasing trend year by year, and the growth of privately owned motor vehicles is more rapid, which leads to a series of traffic problems. The increase of traffic violations and accidents brings hidden dangers to urban traffic and personal safety. Therefore, higher requirements for traffic management can not be realized for all intersections and sections in the city, and all round-the-clock "manpower monitoring" can be carried out. In order to solve this problem, we must use the technology of "machine vision and image processing", set up an intelligent traffic supervision system in the city, and carry out vehicle type detection and identification of vehicles in the past. In this paper, the problem of vehicle classification based on vehicle image is studied. Through multi-view vehicle classification and sparse coding, an effective vehicle recognition method is designed to locate and classify the vehicles in the video and image collected by the camera, as follows: 1) Boundary positioning. On the computer, the collected images are analyzed and processed to obtain the contours information of the vehicles, that is, the length, width and higher data of the vehicles. 2) HOG features and SVM, are used to construct a multi-view classifier to estimate the visual angle of the classified vehicle images. 3) for the vehicle images with different visual angles, sparse coding is used to extract the effective vehicle features. In each perspective, sparse coding is carried out separately. 4) the results of sparse coding and angle estimation from each angle of view are combined, the final vehicle features are designed, and SVM is used to classify the models and judge the models. Compared with the traditional method, the improvement of this paper is to extend the single-scale HOG feature to the multi-scale mode by extending the sparse representation to the vehicle feature representation. The results show that the classification accuracy of sparse coding method is higher than that of traditional word packet method, and the recognition accuracy can be improved by adopting multi-view recognition method. By combining with multi-level classifiers, the classification accuracy of vehicle image can also be improved effectively.
【学位授予单位】：华侨大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U495;TP391.41

【相似文献】