基于光流的人体运动实时检测方法  

Real-Time Detection Method of Human Motion Based on Optical Flow

作　　者：施家栋[1] 王建中[1] 王红茹[1] 

机构地区：[1]北京理工大学爆炸科学与技术国家重点实验室,北京100081

出　　处：《北京理工大学学报》2008年第9期794-797,共4页Transactions of Beijing Institute of Technology

基　　金：国家部委预研项目(06104040)

摘　　要：针对目前广泛使用的光流法计算耗时严重问题,提出了基于差分图像绝对值和(SAD)与光流法相结合的人体运动检测方法.通过计算SAD检测出运动区域,在已确定的运动区域内进行Horn-Schunck光流场计算,准确地计算出人体的运动信息.在后续处理中,应用形态学的闭运算和连通性分析,较完整地分割出人体运动目标.实验结果表明,该方法有效地提高了系统的计算速度,能够实时准确地对人体运动进行检测.Considering the problem that the traditional optical flow algorithm is a time-consuming computation. An integrated method of detecting human motion is proposed based on the sum of absolute differences （SAD） and optical flow. The motion sub-region is detected by the computation of SAD. The optical flow vectors of the object can be obtained by computing optical flow on the motion sub-region. In the post processing, closing and connected components analysis are used for the segmentation of human object. Experimental results showed the effectiveness and robustness of the proposed method.

关 键 词：人体运动 差分图像绝对值(SAD) 光流法 目标分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]