To solve the problem of train driver’s fatigue state detection

an improved active shape model（ ASM）algorithm is proposed and a fatigue judgment mechanism based on support vector machine（ SVM） is designed

realizing the effective detection of the eyes state of train driver during driving. The improved ASM algorithm extracts the local eigenvectors through the multi-scale local binary pattern（ MB-LBP）

improving the robustness in complex light environment. Considering the effect of the time parameters for the judgment of the fatigue state of the driver

the judgment mechanism use two-level SVM structure. As the result shows

the improved ASM algorithm effectively improves the accuracy of the key point location of face

and the judgment mechanism designed in this article achieves 80 % accuracy in the real dataset testing.