基于运动背景的自适应视频对象分割算法

为解决运动背景中视频对象的准确提取，提出一种基于全局运动的自适应视频对象分割算法。基于特征点计算帧间运动，利用最小二乘法计算摄像机仿射参数进行运动补偿，通过二值开闭重建滤波器进行预处理消除噪声；采用改进的分水岭算法将图像标记成不同的灰度区域，以自适应的光流法对分割的对象信息进行评判，从运动背景中分割出前景对象。实验表明，该算法能准确地从运动背景中分割出视频对象，显著地减少了动态前景对象的分割误差，提高了分割质量，可应用于运动目标检测与跟踪。

Adaptive video object segmentation algorithm based on background motion

In order to divide the video objects from movement background,the author first presents a video object segmentation algorithm based on adaptive background motion,calculates inter-frame motion and affine parameters with the least square method using feature point method,and at the same time reconstruct filter preprocessing with a value of two opening and closing.Then the author puts the image tag in to a gray area of different improved watershed algorithm to get information on the object segmentation using optical flow method adaptive to judge the final segmentation of foreground moving objects.The experiment shows this algorithm can accurately segment the video object.