Recognition of Eye States in Real Time Video

A composite algorithm is proposed to identify the opened and shut states of eye in real-time video. Sequentially, each frame of the video is processed first to find and separate the eye regions; then, the accumulation array of circular Hough transform is applied for detecting circles to recognize the opened state of the eye, and the upper eyelid curve is calculated to identify its bending direction to recognize the shut state; after that, if the state cannot be determined yet, judging by the standard deviation of distance between eyelids, the eye state of blinking is determined finally. An accuracy of 94% for eye state detection is obtained in our experiments by using the proposed algorithm. As an important part, the algorithm is going to be utilized in a fatigue warning system for drivers.