【作者】 李乐虎；

【导师】 杨辉；

【作者基本信息】 华东交通大学 ， 电力系统及其自动化， 2009， 硕士

【摘要】 序列图像中运动目标的检测与跟踪是计算机视觉和图像编码技术研究的主要内容,在机器人导航、智能监视系统、医学图象分析及视频图象的压缩和传输,现代化武器作战系统等领域发挥着重大意义。运动目标的检测与跟踪属于底层视觉问题,是运动目标视觉分析的前提与基础。近年来,由于其广阔的发展前景和潜在的市场需求,已备受各国学者的关注。本文分析了运动目标检测与跟踪的国内外研究现状;研究了固定摄像机下多运动目标的检测与跟踪算法;并用TMS320DM642,设计一校园智能视频监控系统。系统也可应用于某些敏感场景,如银行、商店、停车场、高级军事基地等的实时监控。论文的主要工作如下:●首先,了解与课题相关的国内、外研究现状和发展动向,并总结运动目标检测与跟踪的方法。●研究了图像滤波和分割的方法,提出了快速中值滤波方法,有利于提高系统的实时性。●采用连续三帧差分初步检测出目标区域,用最大类间方差阈值法和数学形态学法精确地提取目标区域,运用8点连通对多目标进行标识。●分析了Kalman滤波和预测原理,采用Kalman预测目标轨迹,方法简单,易于硬件实现。目标识别采用了特征模板匹配方法。以Matlab,VC6.0为工具进行算法验证,SEED-VPM642为系统平台,运用TI公司CCS3.1开发工具完成算法的程序设计。更多还原

【Abstract】 Research on Detecting and Tracking of Moving Objects in serial images is a primary content of the Computer Vision and Image Encoding Technology. Now it’s exerting great significance in the domain of Robot Navigation、Intelligent Visual Surveillance、Medicinal Images Analysis、Video Images Condensation and Transmission, Modernization Weapon System. Detecting and Tracking of Moving Objects is subjected to bottom vision problem as well as premise and foundation for analysis of moving objects. Recently it has gained widely attention by scholars in the world.Methods of detecting and tracking and their research status at home and abroad are analyzed in the thesis. Research methods of detecting and tracking under fixed vidicon, then use TMS320DM642 to design a Campus Intelligent Video Surveillance System. Real time supervision on special scenes such as Banks, Stores, Parking Lots, Military Bases etc can be carried by this system as well. The main research work of this paper is presented as follows:●Firstly, analyze research status at home and abroad and summarize algorithms of detecting and tracking.●Research methods of image filter and division. Put forward a rapid median filter algorithm for gray images. Compared with traditional median filter algorithm, it has greatly advanced processing speed.●Based on difference of three sequential frames to detect area of moving objects, then use Ostu and morphology to gain exact objects’s area,finally, employ eight points connectivity to mark their serial numbers.●Analyze the principle of Kalman filter and forecast and adopt Kalman filtering method to forecast moving object’s track. Such a method is simple and prone to hareware realization. Introduce template feature points matching method to achieve object recognition.Using Matlab and VC6.0 tools to validate algorithms; serving SEED-VPM642 as hard platform; introducing TI CCS3.1 sofeware development tools to achieve algorithm design.更多还原