Abstract: To analyse the color image of driver's face taken by CCD camera under obscure conditions, the image noise jamming is eliminated first by erosion in combination with dilation operation, and the threshold processing is used for the grey level differences in the image to select preliminarily the threshold value of grey level. Then, a nonlinear quadratic equation is used to normalize the grey levels so as to intensify the difference among the grey levels and get again the threshold value which is taken as the segmentation threshold between human face and background. By projecting the grey image both horizontally and vertically, the face area can be determined to implement the face locating. Monitoring tests were carried out in different environments and on different roads in accordance to the method proposed, and the results showed that the mean accuracy comes up to 83% under obscure conditions with the advantages of real time and accurate. It benefits judging if a driver is tired in evening.