一种基于梯度方向信息的运动目标检测算法  

A Gradient Direction Based Moving Object Detection

作　　者：孙瑾[1] 顾宏斌[2] 郑吉平[3] 

机构地区：[1]南京航空航天大学信息科学与技术学院,南京210016 [2]南京航空航天大学民航学院,南京210016 [3]清华大学计算机科学与技术系,北京100084

出　　处：《中国图象图形学报》2008年第3期571-579,共9页Journal of Image and Graphics

基　　金：国家高技术研究发展计划(863)项目(2007AA01Z306);国家自然科学基金委员会与中国民用航空总局联合资助项目(60776812);民航总局科技基金资助项目(E9905-MH)

摘　　要：运动目标检测是智能视觉监控系统的基本内容。在对现有算法分析的基础上提出了一种基于梯度方向信息的运动目标检测算法。首先利用方向信息提取视频图像序列中每一帧的边缘梯度图,然后通过改进传统帧差算法,采用uint8数据格式处理含有时间关系的两帧图像以此确定运动目标粗略边界,经运动目标连通域识别,最后结合梯度方向信息准确提取运动目标的完整轮廓。实验结果表明,该算法克服了传统帧差算法不能准确定位目标的缺点,在室内外复杂背景下均能准确地提取完整的目标轮廓。Moving object detection is an important process in intelligent video surveillance systems. On the basis of analyzing existing detection approaches,this paper proposes a detection algorithm based on a gradient directions. For a video image sequence, edge gradient images of each frame are achieved firstly by means of direction. Secondly, the traditional frame difference is improved and it car extract the rough edges of moving objects by processing two temporal correlative frames using uint8 format. Finally,after connective areas of moving objects are obtained,the complete moving object contours are effectively detected using the gradient direction. Experimental results show that, the proposed algorithm overcomes the shortcoming of not correctly detecting moving objects with traditional frame difference and can effectively and accurately extract moving object contour among indoor and outdoor environments with complex backgrounds.

关 键 词：智能视觉监控系统 运动目标检测 帧差 梯度 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]