The Research Of Low-cost Computer-human Interaction System Based On Blinking

Eye detection technology is one of the central problem in machine vision,which has been investigated in recent years.Many applications are derived due to the combination of eye detection technology and human-computer interaction.such as fatigue driving, gaze tracking. High accuracy of positioning in the eye detection system is often complemented by using the high-precision cameras. However, the high price and potential harm to human body make it impossible in some applications. It is necessary to design the system which is low-cost, safe and easy to implement.In our work,we combine the latest technologys to design a new human eye detection system.First, a new detection algorithm is applied to locate the face; and the use of the human eye detection technology is to determine the location of the human eye,which is followed by the tracking of the human eye; Finally, page flip is realized by the blink detection techniques.We introduce methods with better performance and make some comparative analysis among different algorithms to choosing the algorithms that best fit our application.In the literature of face detection technology, many methods are based on Adaboost algorithm,color information and histogram.Compared to other techniques,the Adaboost algorithm attracts extensive attention due to stability, robustness. For this reason, we apply the Adaboost algorithm in our system.Roberto Valenti’s isophote method and the startburst algorithm are widely used methods in the human eye detection technology.In this paper,we make comparative analysis comparisons between these two methods,and the result shows that the former is better under the circumstance of the completion of pupil feature;We have adopted iris features with SIFT(scale invariance) as tracking key points; we exploit the Mean Shift tracking algorithm during the two continuous images tracking to achieve better results; Current blink detection methods are divided into three categories:color distribution-based detection methods, the template matching method and the frame difference method. In the experiment, we find that the color distribution method is vulnerable to illumination, shadows and other environmental factors, and at the same time,is seriously dependent on threshold; the shape of the fixed template matching methods perform poorly in robustness.and although it is robust to environmental changes, the quality of on-line template is not easy to control; Frame difference method can be very effective when the head holds still, but when the head and background moves slightly, the performance will deteriorate.To solve this problem, we propose an algorithm, i.e., the number of connected components in the eye region will be detected in the forward frame to avoid the background impact. The reuslt proves effective.The combination of the above-mentioned algorithm, we design a new, practical eye detection system which is reliable.low cost.Due to high precision and reliability, we can use our eyes to control the function of a predetermined page.