Detecting　region　of　bottom-up　visual　attention　in　videos　has　become　increasingly　popular　due　to　its　wide　applicability.　In　the　meantime,　it　is　a　challenging　problem　that　saliency　only　stimulated　by　low-level　features　and　without　any　prior　knowledge.　In　this　paper,　we　propose　a　pure　bottom-up　method　to　compute　video　attention　which　is　easy　to　implement.　It　involves　three　parts:　Low-level　feature　extraction,　saliency　calculation　and　maps　combination.　The　saliency　of　visual　attention　is　calculated　by　using　Locality-constrained　Linear　Coding　in　multiple　temporal　scales.　Experimental　results　demonstrate　that　the　proposed　method　can　successfully　detect　the　most　salient　regions　which　human　eye　will　typically　focus　on.　©　2017　IEEE.