Precise Eye Gaze Tracking in Video Frames via Combining Particle Filters Method with Genetic Algorithm

Gaze tracking is a key process in human-computer interactions. A basic approach in this field is to predict the position of the pupil in sequential video frames. Particle filter, the most important method which is based on aforementioned approach, has a low precision in predicting the pupil position though guaranteeing expected speed. To solve this problem, in this paper, genetic algorithm (GA) is used in the sampling step of the particle filter method. As a result, in each frame, diversity of the particles required for predicting the pupil position in the next video frame is preserved and at the same time the monotony of them is reduced. To evaluate performance, corresponding to different particle populations, speed and precision of the proposed method and the basic particle filter method in predicting pupil positions in video frames of eye were computed and compared. Results show that the proposed method, compared to the basic particle filter method, tracks the gaze more precisely in a lower time.