Background subtraction algorithms are mainly used to segment some specific moving objects in an image sequences. Within of the action recognition context, these methods may be proper to generate automatically silhouettes of the human actions. In this way, MuHAVi is a human action dataset which provides a small set of manually annotated silhouettes and a large set of multicamera raw video. The purpose of this work is to use a segmentation algorithm to generate automatically the whole dataset of silhouettes for the MuHAVi raw video. The F1-score unit measurement is the selection criterion as for the best method to generate such silhouettes. This paper focuses especially on background subtraction methods that create a statistical model of the background, typically using a mixture of Gaussian. The best-evaluated algorithm can then be used to generate automatically a set of silhouettes.