Driver　fatigue　is　one　of　the　main　reasons　for　traffic　accidents.　Face　images　include　abundant　fatigue　information.　Most　driver　fatigue　detection　methods　extract　local　facial　features.　This　paper　proposes　to　extract　global　features　to　detect　driver　fatigue.　Firstly　global　facial　features　are　extracted　by　PCA.　Secondly,　weak　classifiers　are　constructed　by　decision　trees　based　on　the　PCA　coefficients.　Finally　AdaBoost　algorithm　is　applied　to　select　the　most　critical　features　and　build　a　strong　classifier　for　fatigue　detection.　The　method　is　validated　on　human　subjects　of　different　genders　under　real-life　fatigue　conditions.　The　test　data　includes　thirty　people＇s　4800　images　with　illumination　and　pose　variations.　In　contrast,　fatigue　is　also　detected　by　LDA　classifier　and　global　features　as　a　baseline.　The　average　recognition　rate　of　the　proposed　method　with　no　more　than　138　AdaBoosted　global　features　is　80.19%　which　is　bettor　than　73.21%　achieved　by　the　baseline　using　600　global　features,　and　the　computational　cost　of　the　proposed　method　is　much　lower.　Copyright　©　2009　Binary　Information　Press.