# Feature idees kevin

Issue

Often confused with c.OOK, why?
c.OOK: ZELFDE-A is just twice the movement of c.OOK. Averaged over two parts of sequence might look the same
When hands move close together, hand tracking fails

solution:
Count how many times two wijsvingers touch eachother
but first check how good hands are tracked at the moment both touche:

To better differentiate between signs that use two hands that touch eachoter or signs where one hand remains still and not near the other hand, maybe calculate min distance between both hands, bare in mind we dont use undetected keypoints

Maybe check how much a hand has moved across the frames to detect if it is a dominant hand , could also be used to detect if he uses both hands, so sum of distances of hand between al frames

when we take the distances between both ring fingers, of a interpolated sequence of 4 with preprocessing, we can see that most classes are hard to differentiate and have outliers

doing min max

I took 0.1 distances as being close together the ring fingers, I then took the amount end average=#/len(sequence)
We can see that (2) 3 and 7 are almost never togehter except some outliers.

if we look at the confusion matrix, we can see that c ook and zelfde A are better differentiable, probably because we now have more temporal data

if we use only the temporal count and avg count

we can see that min and max did a decent job, count improved overal score but not in the 2 focused classes