中文摘要：

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。

英文摘要：

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.