Real-Time Human Eye Tracking And Blink Detection

Research on computer vision has been booming since it became popular in the 1970s. More rapid advances in this area were seen in the new 21 century, driven by the always-increasing performance and the cheap cost for the digital image acquisition. In this filed of computer vision, human eye tracking and blink detection has attracted significant attention.Human eye tracking helps to locate the position of human eye, and is the basic foundation of face recognition. Blink detection is important to identify the status of eye, which may have wide application in the fatigue detection and vision interaction.First , we look back to the research course in these two fields, including the tracking approach such as SIFT feature method, MeanShift tracking, also including the blink detection approach such as eye model, frequency model and conditional random field.Second, based on the cascaded adaBoost framework for object detection, this paper proposes the idea of relative coordinate for sample training, and applies this idea into the human eye tracking. We redefine the positive and false samples and features, and accelerate the primitive algorithm to satisfy the real-time calculation.Meanwhile, to settle the unstability of the blink detection, we introduce the phrase of pupil localization, which enhances the robustness of detection.Finally, we combine all the above techniques to design a vision interaction system—EyeMouse.