One dimensional image processing for eye tracking using derivative dynamic time warping

Abstract

In this paper, one dimensional image processing for eye tracking approach is presented using a low cost webcam. It was found that the algorithm is resistant to illumination variation and is able to track eyes of users with and without glasses in multiple orientations. The size and location of the region of interest (ROI), which contains both eyes, is adaptive. Derivative Dynamic Time Warping (DDTW) is chosen as the classifier for this experiment since it can match patterns from one dimension data sequences with varying length. The DDTW is applied to compare distances of the area with an offline template by investigating separately, the row arrays and column arrays for eye tracking. The online one dimensional image processing for eye tracking procedure shows good accuracy and robustness when implemented at 50 frames/s on a 253 GHz Pavilion DV4 HP notebook. Key words: Eye tracking, adaptive region of interest, derivative dynamic time warping, illumination variation resistant, one dimensional image processing.