计算机应用 ›› 2005, Vol. 25 ›› Issue (04): 778-780.DOI: 10.3724/SP.J.1087.2005.0778

• 图形图像处理 • 上一篇    下一篇

多模型鲁棒估计的运动分割

任建峰1,郭雷1,沈玉利1,2   

  1. 1.西北工业大学自动化学院; 2.湛江海洋大学电子工程系
  • 出版日期:2005-04-01 发布日期:2005-04-01
  • 基金资助:

    国家自然科学基金资助项目(60175001)

Motion segmentation based on multi-model robust estimation

REN Jian-feng1,GUO Lei1,SHEN Yu-li1,2   

  1. 1.School of Automation,North Western Polytechnic University,Xi’an Shaanxi 710072,China; 2.Department of Electrical Engineering,ZhanJiang Ocean University,Zhanjiang Guangdong 524025,China
  • Online:2005-04-01 Published:2005-04-01

摘要:

提出了一种基于多模型结构鲁棒估计的运动分割算法。首先对视频处理对象进行基于 四叉树的分裂合并,获取鲁棒估计的初始运动数目以及相应的运动模型的初始参数,然后通过参数估 计,不断更新模型参数,之后通过把每个运动区域和几个运动模型相关联,来同时估计多个运动的区 域,最后通过小物体的运动检测方法检测出小的运动物体,最终达到分割的目的。试验证明该算法取 得了比较明显的结果。

关键词:  视频对象分割, 鲁棒估计, 四叉树的分裂合并

Abstract:

A novel motion segmentation algorithm based on multi-model robust estimation was proprosed. Fisrtly, the split and merge technique based on quad-tree was proposed to obtain the initial motion numbers of robust estimation and corresponding initial parameters of motion model. Then the parameters were updated by using estimating parameters, and the motion models were merged based on the motion consistency. Lastly small motion objects could be detected by using small-object detection techniques. Experimental results show that the method can obtain promising results.

Key words: video object segmentation, robust estimation, split and merge based on quad-tree

中图分类号: