Adaptive background model for moving objects based on PCA

Background modeling and detecting moving objects in scene is a convenient method in many surveillance systems. We propose an approach that is useful in estimating background. In our approach, first each frame is divided to blocks, and blocks in frame sequences sorted to make block series. Finally PCA process applied to these block series. Based on PCA theorem if there is change in block series which means there is not pure background, the main component of block series is comparable to other components of series. By detecting these regions and neglecting it from scene a background modeled. This approach was known as multi block PCA which was used before for detection changes in images and now in this paper we apply it to video sequences adaptively. In this model dimension of database equals to number of frames which made block series. Also our experiments show that this method is robust in change illumination because the model is updated periodically. Moreover computational complexity of the algorithm and accuracy in localizing moving objects could be compared with other fast clustering based background modeling such as Mixture of Gaussian (MoG) and mean shift technique.