Driver　fatigue　is　an　important　factor　in　many　vehicular　accidents.　When　fatigue,　the　frequency　and　time　of　eye　closed　all　increase.　In　this　paper,　we　present　a　novel　solution　for　fatigue　detection　based　on　eye　features.　Firstly　local　binary　pattern　(LBP)　features　of　eye　areas　are　extracted.　Secondly,　weak　classifiers　are　constructed　based　on　decision　trees.　Finally,　AdaBoost　algorithms　are　used　to　extract　the　most　discriminative　features　from　the　LBP　features　and　construct　a　highly　accurate　classifier　for　fatigue　detection.　The　method　is　validated　under　real-life　fatigue　conditions　with　human　subjects　of　different　genders.　The　test　data　includes　1800　images　with　illumination　and　pose　variations　from　thirty　people.　The　experiment　results　show　that　the　average　recognition　rate　of　the　proposed　method　with　no　more　than　90　selected　LBP　features　is　99.39%　which　is　much　bettor　than　74.78%　achieved　by　the　baseline　method　using　600　global　features.　The　proposed　method　has　a　better　performance　at　much　lower　computational　costs.　©　2008　Binary　Information　Press　December　2008.