Detection Of Eyes' State Based On Image

The enhancement of automobile driving security depends on the research of automobile security technology. As one of automobile initiative security technologies, it is essential to monitor the condition of the driver. In this thesis, we focus on how to estimate the gaze of human eye and detect the state of the driver's eyes in real-time with a common camera, so that we can determine whether the driver is at unsafe driving condition.First of all, we research methods of eye features detection, including eye window positioning, eye-lid and eye comers detection, and iris detection. We regard the upper-eyelid as a section of ellipse arc and use the method of Hough transform to detect ellipse. Then we regard the iris as a circle and use the method of sub-pixel detection to detect sub-pixel edge and the method of Hough transform based on gradient orientation and gradient direction to detect circle. The experiment indicates that these methods have the higher estimating precision and more timeliness under the low resolution image condition.Then, we discuss the eye model. Based on the traditional spindly eye model, we ignore the down-eyelid, only retain upper-eyelid, iris and eye corners. And we adopt a simple method to estimate gaze.Finally, we apply the gaze estimation algorithm this paper proposed in a real-time gaze safety examination system. This system can carry on the monitoring of the gaze direction and the sleepy condition real-time.