Abstract

A family of eye tracking methods work with video images analysis. Eye tracking with these methods have many applications in medical science and humanities such as learning and reading disorders, autism disorders, ADHD disorders, human-computer interaction. These methods are fallen into two groups: invasive and noninvasive. In invasive methods, the imaging cameras are attached to individual's head or glasses. In most invasive methods, IR cameras are used for eye tracking. However, in this work, we aim to mount visible range cameras on the head. Then, we utilize a state of the art algorithm to process video images. Then the gaze point of an eye on the processed image is estimated. In order to track eye movements, two low cost cameras (Mini DV Md80 DVR) are used in our work. The cameras have mounted on glasses in such a way that to cause a little discomfort for an individual. One camera has been devised to capture scene video and the other to capture eye movements. In our project, there was no necessity that the cameras to be synchronized with each other. Furthermore, the scene camera image is only used to estimate the gaze point of the eye. In fact, an image processing algorithm is mainly applied to the captured images from an eye. In order to determine pupil center, all the current related works were examined. From amongst the six great works in this field, we have chosen the best work. Image processing algorithm of this work has four main part as the following: 1) preprocessing stage includes elimination of unwanted regions of images, noise reduction and contrast enhancement of images, 2) pupil center estimation from the camera images captured from an eye, 3) calibration stage, and 4) estimation of the gaze point of an eye in scene image. In this paper, eye tracking results are reported on two datasets. The pupil center estimation deviates approximately 15 pixels from real pupil center in average for the first dataset. This dataset is presented by Swirski. This dataset is publicly accessed via internet. The second dataset was collected with the help of our own eye tracking system. Our preliminary results are promising. Of course, the final results on the second dataset will be reported in the paper presentation time. In this project, we were able to design and build an eye tracking system from two low cost cameras which were mounted on the glasses. Then, recently developed image processing algorithm were successfully used to estimate gaze point of the eye. The disadvantages of our basic design are concluded as the following: 1) offline estimation of the gaze point of the eye and 2) two cameras did not synchronize. Based on our good design and implementation results, we will devise and build an online wireless eye tracking system.