Unsupervised Spatio-temporal Segmentation of Moving Objects in Video Sequences
运动目标的自动空时分割算法

The extraction of moving objects is an important and fundamental research topic for many video applications.This paper addresses an unsupervised spatial-temporal segmentation scheme to extract moving objects from video sequences.In temporal segmentation,an outlier rejection(OR) based object detection approach is proposed to extract initial temporal masks,followed by region growing with a distance constraint to compensate initial temporal masks accurately in order to link discontinuous boundaries and fill some holes.In spatial segmentation,watershed segmentation considering the global information improves the accuracy of segmentation in the spatial domain.By using a fusion module,moving objects are extracted.Experiments on various sequences have successfully demonstrated the validity of the proposed scheme.