In　recent　years,　the　incidence　of　traffic　accidents　has　grown　rapidly,　which　has　brought　great　threats　to　people＇s　lives　and　property.　At　the　same　time,　fatigue　driving　detection　and　early　warning　systems　have　become　a　new　research　hotspot.　The　traditional　fatigue　detection　based　on　facial　features　is　dependent　on　digital　image　processing　too　much.　Judging　the　feature　state　requires　a　lot　of　image　processing　work　and　additional　complex　algorithm　support,　such　as　ellipse　fitting.　When　comes　to　judging　whether　driver　wear　glasses,　the　workload　will　increase　greatly,　and　the　results　of　the　final　test　are　unsatisfactory,　which　also　makes　the　image　processing-based　fatigue　detection　method　not　recognized　by　some　researchers.　This　paper　introduces　a　method　based　on　convolutional　neural　network　for　facial　feature　extraction　and　state　determination.　The　fatigue　state　is　judged　by　blending　the　mouth　state　feature　and　the　eye　state　feature,　which　effectively　improves　the　accuracy　and　robustness　of　the　judgment　result.　Depending　on　the　establishment　of　the　facial　key　point　model,　the　efficiency　of　the　mouth　state　judgment　is　greatly　improved,　and　the　problem　of　the　driver　wearing　the　glasses　is　also　cleverly　avoided.　The　experimental　results　show　that　the　method　can　accurately　and　efficiently　complete　the　fatigue　detection　work,　and　the　performance　is　better　than　other　image　processing-based　detection　methods.　©　2018　Association　for　Computing　Machinery.