Research And Application Of Eye Localization Technology

Eye localization and tracking technology is a very popular and challenging topic in computer vision field recently, Innovation of this technique can bring a very broad application prospects, such as eye tracking, fatigue driving judgment, etc. In this paper, under the background of famous herbalist doctor’s teaching and learning system, in order to protect the patients’ privacy and view the patients’ face at the same time, a eye localization and mosaic processing system was designed and implemented. This is a new application of this technology.It is very challenging to detect and track eye correctly, because is will be affected by the background and various factors such as eye gesture. This paper considers the influence of illumination change, eye gesture deflection and many other factors, constructing a multi-scale eye localization and tracking system of single object in complex background.After extracting video frame, using color correction algorithm to improve the accuracy of color extraction, then do the color extraction, morphological operation and candidate face region screening one after another. In these areas remained after above steps, we convert color image to grayscale and do the light compensation. Finally the face region is obtained by Ada Boost face detection algorithm. Next we use the Ada Boost eye detection algorithm to obtain eye region, extracting feature points in this area, then we use the improved KLT method to track feature points and calculate the size and location of the human eye area, adding a mosaic window in eye area to protect user’s privacy. In this paper, the main work is as follows:1. Put forward an adaptive elliptical skin color model, this algorithm first train samples in YCb Cr color space, getting offline elliptical skin color model, this algorithm can adapt to the illumination change of the scene according to the judging standard of original offline model parameter, this algorithm solve the problem of interference of skin-like color background;2. Put forward using skin color model based on motion tracking in video stream environment, compared with previous use of static skin color model, this model take full use of human movement information in video, segmentation effect and real-time performance is good, and it can filter out still large skin color like region of background.3. Put forward two kinds of improved measures of face detection in simple background andcompex background respectively, using experimental results indicate that these two improved methods enhance the performance of face detection in a certain extent.4. In order to solve the CamShift failures because of the hsu similarity of eye area and skin, we take advantage of the hsu uniqueness of pupil in face to initalize search window with pupil region, the robustness of the human eye tracking system is improved.The eye localization and mosaic processing system designed and implemented in this paper has good real-time capability, robust to a certain range of illumination variation, has certain practical significance.