视频运动目标检测方法的对比分析  

Comparative Study on Detection Methods for Video Motion Targets

作　　者：赵文哲[1] 秦世引[1] 

机构地区：[1]北京航空航天大学自动化科学与电气工程学院,北京100191

出　　处：《科技导报》2009年第10期64-70,共7页Science & Technology Review

基　　金：中国高技术研究发展计划(863计划)项目(2006AA04Z207);国家自然科学基金项目(60875072);中国科技部国际科技合作项目(2007DFA11530);中国教育部高等学校博士学科点专项科研基金项目(20060006018)

摘　　要：针对视频运动目标检测领域的研究现状,对帧间差分法、背景建模法、图像分割法、聚类分析法、运动矢量场法等视频运动目标检测方法的算法适应性、复杂度等性能进行了对比分析。辅以典型的视频运动目标检测实例,比较了各种方法的适用范围与优劣特点,可为不同应用场合下视频运动目标检测方法的选择提供参考依据。指出了现阶段视频运动目标检测技术研发的瓶颈所在,以及在新的应用需求背景下所面临的挑战,并对进一步的发展趋势作了预测和评估。This paper reviews the current development of detection methods for video motion targets, with a comparison of a series of related algorithms, including frame difference method, background subtract method, background modeling, method based on image segmentation, motion vector field analysis method and others, together with their application ranges and complexity. Performances of several target detection methods are tested with video samples, based on which the accuracy, the level of real-time and independency are compared to indicate their applicability and limitations. Moreover, some important issues in engineering applications and possible solutions are proposed based on the current development trends and requirements of practical applications.

关 键 词：目标检测 帧间差分法 背景建模法 AdaBoost分类 运动矢量场 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]