Vibration-resisted video segmentation algorithm based on hierarchical Markov random field model

CHU Yi-ping,Xiuzi Ye,Yin Zhang,Sanyuan Zhang

DOI: https://doi.org/10.3785/j.issn.1008-973x.2007.11.005

2007-01-01

Abstract:A new video segmentation algorithm for data captured with possible camera vibrations was proposed.The algorithm hierarchically models the frames of video sequences via hierarchical Markov fandom fields(MRF),and achieves accurate segmentation of objects under camera vibrations by using spatial neighboring relationships in each frame.Gibbs sampling was used to obtain maximum A posteriori(MAP).Experiments on video sequence data in the environments of strong illuminations,multiple objects and complex backgrounds show that new algorithm is more efficient than the existing algorithms based on backgrounds accumulated from frame difference and based on Gaussian mixture models(GMM).