动态场景视频序列中的前景区域自动提取

首先提出一种具有自适应外点过滤功能的全局运动估计算法，通过交替地进行参数估计和外点过滤，能够有效地抑制噪声的影响，实现准确的背景对准；然后，通过直方图拟合获得准确的背景噪声方差值，克服了以往只能依据经验设定背景噪声方差的缺点，并使用显著性测试技术有效地对帧差图进行二值化，最终消除帧间的重叠背景，提取出前景区域．实验结果表明，该算法能够有效地实现动态场景视频序列中的背景对准，并准确地提取出前景区域．

Automatic Extraction of Foreground Areas in Video Sequences of A Dynamic Scene

The work in this paper investigates the automatic extraction method of foreground areas in video sequences of a dynamic scene. In the method we proposed, firstly, a robust global motion estimation algorithm is proposed with ability to filter outliers adaptively. By estimating the global motion parameters and filtering the outliers alternately, the noise is restrained and the background is aligned accurately. To overcome the shortcoming of setting the value of the background noise variance by experience, we estimate the value by histogram fitting. Then, we use the significance test to threshold the difference image and extract the foreground areas. Experiments demonstrate that our method can align the backgrounds in successive frames and extract the foreground areas effectively.