Real-time driver's eye state detection

Abstract

Driver's eye state is a key factor for identifying the driver's visual attention. This paper proposes a real-time system to detect the driver's eye state. The system combines several basic image processing techniques which can achieve better performance than other single arithmetic. And we contrive different methods to detect the eye state in the daytime and at night. Firstly, the system combines the skin-color matching & vertical projection in the daytime and applies the intensity adjusting & vertical projection at night to detect and track the face region, and then validates it by geometric symmetry of face; secondly, we apply the horizontal projection to detect the eye position and validate it by some facial features; Lastly, a new complexity function is employed to judge the eye state, and the frequency of eye closure is registered. The method in our paper can detect whether the eyes are close more efficient than many other methods and it is more real-time with less computation. The experiment results show that this method is a robust and real-time one.