The Human-Computer Interaction System Based On Eye Tracking

At this stage, eye tracking is an important technology in human-computer interactionsystem, which has extensive application prospect in vehicle, psychology, military andmedical. Many methods had been applied in eye tracking such as intrusive mood andnon-intrusive mood. Because intrusive mood is cumbersome to use, study in non-intrusiveis more popular which had achieved results. The existing problems in eye tracking areuser-friendly, operating efficiency, real-time performance, accuracy and stability.In this study, we use the non-intrusive eye tracking which depends on no externalequipment. It composed of face and eye detection, image preprocessing, pupil centercoordinate extraction, inner corner extraction, eyeball space modeling, gaze estimationmodeling, eyes space and computer screen space mapping function.Firstly, we use a improved bilateral filtering to preprocessing images gathered fromcamera. This improved bilateral filtering reduces dimension of wighted coefficient by oneand is combined with median filtering. After the preprocessing of images, gaussian noiseand pepper noise are forbidden. Secondly, Histogram equalization is used to increasecontrast which increase the scope. Then, a Adaboost function based on Haar feature is usedto face and eye detection. We use a improved hybrid projection function to roughly locatepupil center according to gray level distribution. Then we can use the pupil center asthreshold to binervate eye image. We will get a accurate pupil center by circle synthesis.Because of the unconsciously movement of human eyes, we take the average of three pupilcenter as a finally characteristic parameter, then a local image between pupil center and thecorner of eye is used as a smaller corner detection scope. During the processing, a improvedSUSAN function is used to finish corner detection which using a self-adaption threshold.Since then, we have got all the characteristic parameters. We presents a simple eyethree-dimensional space model according to the structure characteristics of eye trackingwhich can reduce the errors in projecting3D to2D. This model is a mapping fromparameters gathered from eyes to computer screen coordinates. We can get more accurateresults by adding a error compensation parameter. Finally, we can judge the mouse move orclick according to the time staring at some area.Here, we build a human-computer interaction system based on the above theory whichcan real-timely find eye gaze under natural light environment. VS2005is the developmentplatform, and a USB camera is the only external equipment. It has lower requirements and easy employment compared with other mainstream eye gaze tracking methods.