摘要
提出一种多尺度帧间边界变化检测方法,将当前图像划分为变化区域和非变化区域,变化区域内采用基于Hausdorff距离跟踪器找到对象在后继帧的最佳匹配位置;然后利用Snake模型拟合该位置上的非刚性形变,得到对象真实边缘;最后采用一种基于距离变换的最短路径法使开环闭合。
A method for multi-scale change detection is proposed based on inter-frame edge difference. The result shows that the algorithm is robust and well sensitive to change. Snake model is used to extract the object contour. Because of blindness of Snake model, Hausdorff tracker is introduced prior to snake tracker. Snake model is used to track non-rigid deforma- tion in object translation position indicated by Hausdorff tracker. The result of snake model is not always closed, a method of the shortest path based on distance transform is proposed. The segmentation algorithm avoids many of the complex problems associated with optical flow estimation and gray-based segmentation in the space. Experiments also demonstrate this algorithm can handle many traditional types of sequences.
出处
《武汉大学学报（信息科学版）》
EI
CSCD
北大核心
2006年第8期748-751,共4页
Geomatics and Information Science of Wuhan University
基金
国家973计划资助项目(2004CB318206)
关键词
视频对象分割
对象跟踪
活动轮廓
距离变换
video object segment
object tracking
Snake model
distance transform
