Extended PCA and SGM based algorithm for foreground segmentation

Abstract

A real-time background segmentation system has already proposed based on two methods Principle Components Analysis (PCA) and Single Gaussian Model (SGM). But PCA based method is required to calculate of decomposition and for this reasons time consumption is vast. On the other hand SGM based foreground segmentation model which is reliably deals with lighting changes, detects repetitive motion from clutter and long-term scene changes. This paper presents a method consists with two methods extended PCA (EPCA) and SGM. An adaptive strategy is used to integrate two methods. We adopt the adaptively incremental eigenspace model to build the intensity information for each pixel. Weng designed Candidate Covariance-Free Incremental PCA (CCIPCA) method, which does not need decomposition. Moreover, it is also simpler and faster than other proposed incremental PCA (IPCA) methods. Shadow removal using adaptive filter and post-processing techniques are used in a proposed method. Therefore, as the proposed algorithm does not require decomposition so this method is faster with PCA and SGM based foreground extraction method. Experimental results represent that this proposed model is robust to noise and illumination change due to inheriting eigenbackground and Gaussian's model advantages to improve the segmentation results.