The　3D　facial　surface　demonstrates　rich　information　about　human　beings＇　expressions.　However,　methods　to　recognize　humans＇　facial　expression　are　mainly　still　focusing　on　2D　images,　which　is　not　robust　to　pose　and　lighting　conditions.　In　this　paper,　the　problem　of　the　person-independent　facial　expression　recognition　is　addressed　on　basis　of　the　line　segments　connected　by　specific　3D　automatically　detected　facial　keypoints　and　LBP　features　of　depth　images　around　the　automatically　detected　facial　keypoints.　Using　a　Support　Vector　Machine　classifier,　the　recognition　rate　reaches　up　to　92.1%　on　the　BU-3DFE　database.　Comparative　analysis　shows　that　our　method　outperforms　the　competitor　approaches　using　similar　experimental　settings,　which　proves　the　effectiveness　of　our　method　for　3D　facial　expression　recognition.　©　2016　IEEE.