关键词: 驾驶疲劳识别, 疲劳表情, 正交流形保持投影, 信息融合, 贝叶斯网络

Abstract: In order to surmount the limitations of Locality Preserving Projections method（LPP）, Orthogonal Manifold Preserving Projections（OMPP） method is proposed, which preserves the local and global structure invariance of the samples in the low-dimensional space by integrated non-adjacent constraint information on the objective function of LPP. The orthogonalization process is applied to solving the projection matrix for further reducing the characteristics dimensions after projections, and all of these improve the accuracy of driver fatigue recognition by the expression. In order to further reduce the false alarm rate of recognition，the Bayesian net is applied to detecting the fatigue states by the fusion of multiple features such as fatigue expression, frequency of yawn, and the degree of eye closure. The superiority of the above process has been verified by experiments.

Key words: driver fatigue recognition, fatigue expression, Orthogonal Manifold Preserving Projections（OMPP） method, information fusion, Bayesian net