The Research And Implementation Of Eye Tracking Technology

Eyes are the first channel to access the outside world, and it is important to study how to use the characteristics of human eyes to get the wanted information. At present, the related research of eye tracking technology is a hot research direction in the fields of computer vision and human-computer interaction. Eye tracking technology has a very wide range of application prospects, including military aviation, biomedicine,human-computer interaction, real-time fatigue driving motor vehicle inspection and so on.In this paper, we mainly study the face detection, facial feature points detection, eye area extraction, location of pupil center, sight direction estimation and so on, of which the pupil center location is the key study content, and on the basis of predecessors’ research, we propose an improved pupil center location algorithm which is based on image gradient. This method can well eliminate the interferences of eyebrow area and hair area which have the similar color of pupil area, so this method can find the eye center more accurately even in low resolution images,and can satisfy the requirement of real-time.Firstly, we detect face and facial feature points in the given picture.After summarizing the existing common face detection and feature points detection methods’ advantages and disadvantages, we choose the adaboost algorithm to detect face, and use deformable model fitting method to detect facial feature points. By using this method we can get 66 facial feature points. Each eye includes 6 feature points and each eyebrow has5 points. By extracting these feature points we can get the eye area accurately. Experiments show that this method can satisfy the requirementof real-time, and can eliminate the interference areas such as eyebrows,and in the small range cornering head situation and the looking up or down situation this method can still get eye area accurately. And use the detected feature points we can preliminarily estimate gaze direction.We perform a series of image processing to the eye area image to generate the eye weight map image, then taking advantage of this weight map image we get the improved pupil center location algorithm which is based on image gradient. We use both the eye weight map image and image gradient in the pupil center location algorithm. Using the weight map image we can eliminate the influences of such dark areas as eye corners or eyebrows, which can improve the accuracy of pupil location. We use eye image to generate weight map image. We perform a series of image processing to the eye area image, for example, image binaryzation, morphological operation, then we will get the weight map image, which has higher value in pupil area. In our improved algorithm, we combine the weight map image with image gradient, then do the calculation of image gradient and displacement vector, and screen out the points that wouldn’t be the pupil center point obviously, and then calculate the average value, then the point which has the biggest value will be estimated as the position of the pupil center. Then we do the preliminary gaze direction judgment.Traditional gradient image pupil center location method traverses all the pixels in the image, and often mistakenly locate to the eyebrows and eye corners. The experimental results show that our improved algorithm can well exclude the interferences of eyebrows, hair, and eye corner area,and can find the pupil center position more accurately.