Detecting　driver　fatigue　is　extremely　important　to　improving　transportation　safety.　To　account　for　the　temporal　aspect　of　human　fatigue,　we　propose　a　novel　approach　based　on　coded　dynamic　features　to　detect　fatigue　in　facial　image　sequences.　First,　global　features　are　extracted　from　each　facial　image　and　concatenated　into　dynamic　features.　Then,　we　encode　the　dynamic　features　into　binary　strings,　and　construct　weak　classifiers　on　them.　Finally,　statistical　learning　algorithm　is　used　to　extract　the　most　discriminative　features　from　the　pool　of　coded　dynamic　features　and　construct　an　accurate　classifier　for　fatigue　detection.　The　proposed　approach　is　validated　on　real-life　fatigue　data　with　illumination　and　pose　variations.　Experimental　results　show　the　superiority　of　the　proposed　approach　over　the　baselines　and　an　encouraging　correct　rate　is　achieved.　1548-7741/Copyright　©　2008　Binary　Information　Press　October　2008.