Аннотация:Almost any recording of eye movements is noisy. In most cases, this noise can
be neglected, since the error introduced is quite small. But when studying micro-movements
of the eyes, the noise level of the equipment matches the amplitude of the useful
signal, so it is necessary to smooth the data. Typically, linear smoothers are used to reduce the noise. Since they are very sensitive to outliers and leaps, preliminary median smoothing
is performed. It would be useful to have one robust procedure as an alternative to this
two-step smoothing process. In this work, another technique is suggested, based on the Huber
M-smoother (Tsybakov, Doubrovski, unpublished report, 1990). A program implemented
in the MATLAB / OCTAVE environment allowed us to calculate quickly Huber’s estimates in
a finite number of steps. The smoothing algorithm was tested on eye movement data containing
micro-saccades with amplitudes up to 1 °, and showed high efficiency.