Adjacent and functional LBP based background model learning for video object detection

In this paper, we have proposed two variants of LBPs and a new model learning scheme for efficient background modeling and hence foreground detection. The first one is known as adjacent LBP (LBP-A) while the second one is called the functional LBP (FLBP). The histograms of these two proposed LBPs have been used to model the background independently. These background models are updated through learning the new information of every incoming video frame. We have proposed a new learning strategy for updating the model histograms. A new weight adaptation strategy has also been proposed based on the notion of proximity. The proposed algorithm has successfully been tested with video sequences from PETS database and the performance has been compared with that of GMM and LBP based schemes [3].