C. Validation Gate
A validation gate is implemented to remove outliers.
Region matching by normalized cross correlation suffer from
the aperture problem, which occurs when the values of the
pixels does not change in the direction of the motion [14]. In
these situations the similarity image will provide a bank with
exactly the same values on the top. One way to determine the
most probable match is to create a score function based on
distance to previous match and peak height. In the case of a
bank with lots of point with similar values to the point on the
top closest to previous match will be selected. If no peak is
found at all, the validation gate selects the point with highest
value in the similarity image.
There are usually several peaks in the similarity image
when an outlier occurs. If all peaks where selected for
evaluation even a small peak at the same point as previous
match will get a high score, since the distance is zero.
Therefore only peaks within a certain range from the
maximum peak value are selected.