摄像机运动的视频图像分割  

SEGMENTING MOVING VIDEO IMAGE OF VIDEO CAMERA

作　　者：宋春玉[1] 

机构地区：[1]黑龙江科技学院计算机与信息工程学院,黑龙江哈尔滨150027

出　　处：《计算机应用与软件》2010年第9期262-264,285,共4页Computer Applications and Software

摘　　要：通常所研究的视频图像序列是由摄像机对真实场景的拍摄形成的。在拍摄过程中摄像器件会发生缩放、旋转、平移等运动,当视频中的分割目标也在运动时,就会导致运动目标在图像上造成的变化与背景本身的变化混淆在一起,很难区分出哪个是前景,哪个是背景。在这种情况下如何精确地实时分割出运动目标成为研究的重点。采用时空联合的分割方法,先对视频图像进行空域单帧分割,采用形态学重建对图像进行处理,通过分水岭算法,得到精确的分割效果;根据图像序列间的运动信息,采用背景运动估计和补偿技术,以6参数仿射模型为运动模型,通过超松弛迭代获得仿射模型参数,取得了较准确的运动估计结果。通过全局运动补偿,对当前帧与补偿帧进行差分运算显露出局部运动区域。在此基础上对已有的分割结果进行区域合并,分割出运动目标。实验证明,本算法能准确实时地分割出运动目标。Usually, the video image sequence studied is formed from real scene shooting by video camera. During the shooting process, there often occurs the motion of the video device including zooming, rotation and translation, etc. , when the target in video to be segmented is also moving, it will lead to the mixing of the changes on image incurred from the moving target and the changes of background its own, so it is diffi- cult to differentiate the foreground or the background. In this case, how to precisely extract moving target timely has become the focus of the study. We used spatio-temporal segmentation methods in this paper. First we made single frame segmentation on video image in spatial do- main,and used the morphological reconstruction to process the image,with the watershed algorithm we got an accurate segmentation;Second, according to the motion information in image sequence and utilising background motion estimation and compensation techniques,we took 6-co- efficient affine model as the motion model, parameters of the affine model was acquired through over-relaxation iteration, thus the fairly accu- rate motion estimation results were achieved. By global motion compensation to make differential operation on current frame and the frame of compensation to unveil local motion region. On this basis, we made the combination of the regions with segmented result to segment moving target. Experiment proves that the method can precisely and timely segment moving targets.

关 键 词：形态学重建 分水岭 超松弛迭代 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]