Driver　fatigue　is　one　of　the　main　reasons　causing　traffic　accidents.　Yawning　is　an　important　character　of　driver　fatigue.　Mouth　geometric　character　represents　state　of　driver.　So　driver＇s　mouth　detection　and　information　extraction　is　especially　important.　This　paper　proposes　to　locate　and　track　driver＇s　mouth　using　two　CCD　cameras　to　detect　driver＇s　fatigue　in　real-time.　One　of　them　is　with　cloud　platform　control.　Camera　A,　which　is　fixed,　is　mainly　used　to　supply　driver＇s　head　position.　Camera　B,　which　is　controllable,　is　to　locate　driver＇s　mouth　and　extract　mouth　information.　First,　collecting　monitoring　video　of　camera　A　(low　resolution　ratio)　and　locate　driver＇s　face　using　fast　image　processing　algorithm.　Second,　we　send　commands　to　camera　B　(high　resolution　ratio)　through　serial　port　to　make　it　shoot　driver＇s　head　all　the　time　according　to　face　position　in　Camera　A.　Then　we　use　haar-like　features　to　detect　driver＇s　mouth　in　camera　B　and　track　it　according　to　historical　position.　At　last,　yawning　is　detected　by　the　ratio　of　mouth　height　and　width.　Through　this　method　the　resolution　ratio　of　driver＇s　mouth　is　higher　than　one　camera　and　the　feature　information　is　more　accuracy.　It　provides　a　better　basis　for　driver　fatigue　judging.