First it says to choose the width l of the interval of points used to decided whether (t,y(t)) is an extreme.
Now later in the paper it says $y(t)$ is a local maximum only if
$$ \ y (t - j) < y(t) > y (t+ j) $$ for 0 < j < l
It seems this is not a method for finding local maximum in real time as they occur because you would only know if y(t) was a maximum for the interval 0 < j < 10 after the fact.
Is my interpretation correct?