基于块仿射分类和HD跟踪的视频分割方法*

提出一种自动视频分割方法,分为运动对象检测、对象跟踪、模型更新、分水岭轮廓提取四个阶段。与变化检测方法不同,该基于块的运动分类器能够检测背景具有一致运动情况下的运动对象。自动得到运动对象的二值模型并在随后帧中使用Hausdorff距离进行跟踪。将视频对象运动分为慢变和快变两部分,分别结合背景边缘模型进行匹配更新。最后提出彩色多尺度梯度修正的分水岭算法提取对象的轮廓。实验证明了算法的有效性。

Video segmentation method based on block affine classification and Hausdorff tracking

A new video segmentation method which consisted of motion object detection,object tracking,model update and contour extraction by watershed was proposed.Compared with the change detection methods,the proposed block-based affine motion classifier allew coherent moving background.Binary object models were derived from edge information.The derived models were then tracked in the subsequent frames using Hausdorff object trackers.Both rapid change and slow change component of moving object were updated combined with background edge model.At last a proposed color multi-scale modified watershed-based technique ensured the accuracy of the new model boundary.Experimental results demonstrate the performance of the algorithm.