摘要
在动态场景中提取运动目标是开展视频分析的关键问题,也是当前计算机视觉与图像处理技术领域中的热门课题。本文提出了一种适用于动态场景的运动目标提取新算法,算法先根据摄像机全局运动模型计算全局运动参数,再利用三帧差分法得到分割的前景。将分割为背景的像素点映射到邻近帧,求得各帧的像素点为背景时其高斯模型的均值及方差。最后利用粒子滤波预测出下一帧前景区域,计算各像素点为前景的概率,获得运动目标的视频分割结果。实验表明,本文算法有效地克服了由于全局运动模型参数估算偏差而导致的累积误差,能以更高精度实现跳水运动视频中的目标分割。
Moving object extraction in the video sequences of a dynamic scene is a key problem to video analysis, and also a hot research topic in computer vision and image processing. In this paper, a new extraction algorithm for moving objects in the video sequences of a dynamic scene is proposd. First of all, by using the computed global camera motion models to map the pixels in two consecutive frames, we obtain the coordinates of the pixels. Afterwards, by mapping the pixels which are used as the background of segmentation to the nearby frames, we get a mean variance of those pixels in the Gaussian model. Then,we use particle filtering to foresee an anticipated foreground district for a coming frame. Thus, we can compute the probability for all possible coming pixels, and obtain the segmentation results. Experiments show that the algorithm can effectively reduce the accumulating error caused by inaccurately estimating the parameters of the global camera motion models,and precisely segment the moving objects in the diving practice video.
出处
《计算机工程与科学》
CSCD
北大核心
2009年第4期32-35,共4页
Computer Engineering & Science
基金
国家自然科学基金资助项目(60673093
60573079)
长江学者和创新团队发展计划资助项目
怀化学院计算机应用技术重点学科资助项目
关键词
视频运动分析
目标分割
粒子滤波
motion analysis in video
object segmentation
particle filtering
