Fatigue　may　lead　to　potential　accidents,　but　its　diagnosis　is　difficult　so　it　is　easy　to　be　delayed　or　missed.　In　this　paper,　a　novel　3D　facial-image-based　fatigue　detection　method　is　presented.　There　are　three　steps　involved:　First,　3D　surface　curvature-based　methods　are　combined　with　some　2D　image-based　methods　to　locate　the　position　of　facial　fatigue　feature　points　on　the　facial　3D　image-based　model,　which　is　reconstructed　based　on　a　pair　of　binocular　facial　images.　Secondly,　facial　fatigue　features,　such　as　eye　blink,　gaze　direction,　mouth　morphology,　and　facial　expressions　are　extracted　from　the　facial　3D　image-based　model　by　the　corresponding　extraction　methods.　Finally,　together　with　N-times　sampling　data,　the　attribute　value　of　each　facial　fatigue　feature　is　calculated,　and　then　integrated　together　to　recognize　fatigue　by　linear　discriminant　analysis　(LDA)　algorithm.　The　following　experimental　results　show　that　compared　with　only　2D　image-based　method,　the　accurate　location　of　facial　feature　points,　attribute　value＇s　calculation,　and　fatigue＇s　fusion　detection　are　obviously　improved　using　the　method　presented　in　this　paper.