Abstract: An improved segmentation algorithm of moving objects based on background reconstruction is proposed.Firstly the coarse moving objects are separated from the background based on the frame difference.The combination is applied to some conjunct parts from the subtraction.The morphologic dilation is operated to remove the hole in the mask.And then the background is rebuilt and accumulated according to the original moving object mask.Lastly according to the background subtraction and the frame difference，the judge principle is decided and the precious moving object mask is obtained.Experiments demonstrate that the proposed method is effective even if the moving objects are large or the camera’s orientation is parallel to the moving objects.

Key words: background reconstruction, moving object segmentation, background subtraction, frame difference

摘要： 提出了基于背景重建的运动目标分割改进算法。首先使用多帧差获得初次的背景和前景分离，使用前后多帧差重合部分的合并作为当前帧的运动对象，并通过形态膨胀操作消除二值模板中的空洞，得到该帧的并不精确的初次运动目标掩膜。通过掩膜获得稳定的累积背景。通过得到的背景差和帧间差相结合制定判断原则，获得精确的运动目标掩膜，从而对运动目标进行分割。实验表明该算法对于较大面积的运动目标，或者当摄像机拍摄方向与目标运动方向一致的情形下都可以得到较准确的效果。

关键词: 背景重建, 运动目标分割, 背景差, 帧间差