Performance comparison of signal processing filters on smooth pursuit eye movements

Noise that appears during eye movements data recording can cause inaccuracy in data readout. Various signal processing filters can be used to remove this noise, particularly during smooth pursuit eye movements. However, performance comparison of those signal processing filters is yet to be known when they are implemented in a smooth pursuit-based calibration method. In this study, we compared three signal processing filters namely Moving Average, Gaussian and Kalman filters to remove noises in smooth pursuit eye movements. In the experiment, we compared the performance of Moving Average, Gaussian, and Kalman filters. From the experimental results, Moving Average filter yielded errors of 36.97 ± 10.62 pixel (horizontal position) and 48.07 ± 15.11 pixel (vertical position). Gaussian filter yielded errors of 37.74 ± 11.23 pixel (horizontal position) and 51.06 ± 17.62 pixel (vertical position). Kalman filter yielded errors of 56.06 ± 30.97 pixel (horizontal position) and 72.98 ± 41.21 pixel (vertical position). Experimental results show that Moving Average filter yielded the best accuracy compared with the other signal processing filters. In future, our results maybe used in development of unobtrusive calibration procedure for spontaneous gaze-based interaction.