Recognition of driver eyes' states based on variance projections function

The detection of drivers' eyes states is a key part in detecting drivers' vigilance states based on vision analysis. The present methods were not in real-time due to complex algorithm, or were very sensitive to the variety of illumination, and they are not suit to drivers work condition. We presented a new method which is based on variance projections function. Variance projection reflects the variance of intensity along projection direction mainly. The intensity of eyes area changes orderly under any illumination. So the eyes' feature can be extracted through variance projection at any illumination. A lot of eyes area variance projections under different illumination have been studied and we find out there were distinct differences between the projections when the eyes were open and when the eyes were close, so we can recognize the two states of the eyes using these features. The experiments result that the method is Robust to different illumination condition.