A CNN based real-time eye tracker for web mining applications

Abstract

Eye gaze tracking is an increasingly important technology in the field of human-computer interaction. Individuals’ preferences, tendencies, and attention can be measured by processing the data obtained from face and eye images. This technology is used in advertising, market research, web page design, education, learning methods, and various neurological-psychiatric studies of medical research. Many different methods have been used in eye gaze tracking tasks. Today, commonly model-shape and appearance-based methods are used. Model-shape based methods require less workload than appearance-based methods. But it is more sensitive to environmental conditions. Appearance-based methods require powerful hardware, but they are less susceptible to environmental conditions. Developments in technology have paved the way for applying appearance-based models in eye gaze tracking. In this paper, a CNN-based real-time eye tracking system was designed to overcome environmental problems in eye gaze tracking. The designed system is used to determine the areas of interest of the user in web pages. The performance of the designed CNN-based system is evaluated during the training and testing phases. In the training phase, the difference between the desired and determined points on the screen is 32 pixels and in testing phase, the difference between the desired and determined points on the screen is 53 pixels. The results of the test trials have shown that the proposed system could be used successfully in eye tracking studies on web pages.