Fast robust foreground-background segmentation based on variable rate codebook method in Bayesian framework for detecting objects of interest
Zhao Yongjia, Liu Weihua · 2014
In this paper, a reliable pixel-based foreground-background segmentation technique for detecting object(s) of interest (OOI) from video sequence captured by a fixed camera is proposed. OOIs, used for further tracking or positioning applications, should be detected accurately from those moving (or still) objects even under variable illumination and the corresponding background model need to update quickly. To cope with fast change of the scene, we present an adaptive variable rate codebook update algorithm based on the cache mechanism, which adjusts the time thresholds according to the number of current effective samples in codebook. Then Bayes rule is employed to make the final decision based on prebuilt OOIs' color model and the background model deduced from the codebook. The experiment results have proven the given method's effectiveness.