In the training mode, the algorithm uses default parameter
settings (e.g., predetermined averaged blink duration,
slope sizes, interblink duration) to perform eyeblink
detection. The algorithm then does a simple
statistical analysis (mean and variance) of the eyeblink
detection results. The algorithm automatically adjusts
the detection parameters to provide a better fit ofthe eyeblink
characteristics of the subject under study. The algorithm
then repeats the eyeblink detection (with the
new parameters) and continues the parameter updating
process until the parameters converge or the number of
iterations exceeds a predetermined value. Our experience
suggests that the learning process converges in
fewer than five iterations, and the entire training process
is completed without any operator intervention