Abstract: A novel method of gaze point estimation used in single camera system， which has faster training speed and is suitable for multiple classifications， was proposed based on ELM. In the initial calibration phase， multi gaze parameters served as ELM input， while the gaze point area was ELM output. The nonlinear polynomial was used as activation function. ELM training data was obtained through the initial calibration， and then the mapping model between the line of sight parameters and the gaze point was established. Through estimation of gaze point distribution of different angles and changing the number of hidden neurons， it was found that the accuracy and stability of gaze point obtained by ELM method are better than those obtained by traditional nonlinear polynomial model.

Key words: gaze tracking, ELM(extreme learning machine), gaze point estimation, mapping model, polynomial model