Saccade derivative filters and their clinical implications.

A saccade model is used to evaluate the accuracy of three derivative algorithms, the two-point central difference derivative filter, band-limited derivative filter and median derivative filter. Noise contaminated saccade simulations are used to compare the known velocity and acceleration waveforms with estimates from each of the three algorithms. The median derivative filter, a nonlinear differentiating digital filter, provides the best velocity estimates from an eye position signal. This filter has no impulse response function, eliminates Gibb's phenomena evident in all linear filters, and is easy to implement.