Video object segmentation algorithm based on global motion estimation
Zhang Fan · Jisuanji gongcheng yu sheji · 2007
Video's global motion(camera motion) behaves time-relativity of video sequence.It have more higher semantic information than any other video's character.In order to acquire video's global motion more efficient,a novel method for video's global motion es-timation based on singular value decomposition(SVD) is presented.Firstly,block matching is used to obtain local motion field,and then the global motion parameters is estimated by the singular value decomposition(SVD) of matrix.Then the coarse mask of foreground object is obtained by using morphological motion filtering.At last,using this mask with the information of edge synthetically to segment the moving object.Experiments show that the moving objects of video sequence with global motion character can be extracted out using the method.