The invention discloses a multi-feature fusion driver abnormal expression recognition method. The multi-feature fusion driver abnormal expression recognition method comprises the steps: S1, tracking and monitoring expression actions of a driver in real time through a camera installed on the driver side; S2, precisely identifying expression details in the real-time driver video; S3, detecting the positions of the eyes, and judging whether the eyes are tired or not; S4, positioning the edge contour of the mouth, and judging whether yawn action occurs or not; S5, detecting a head action, and judging whether fatigue occurs or not; S6, weighting the detection results of the eye state, the mouth state and the head motion state, finally judging whether fatigue occurs or not, and outputting the detection results; and S7, combining the identification result of the current frame as an estimated position of subsequent frame identification, and respectively detecting actions in subsequent frames to realize continuous detection and identification of abnormal behaviors of the driver. According to the invention, real-time monitoring and alarm triggering can be carried out, a driver is warned andreminded, and traffic accidents are prevented, and the safety in the driving process is ensured.