Gon_FREEKS:
as the speed increases the pulses are getting closer, and the pulse width shorter.
That's great, it means that you can rely on ratios between pulse length and time to detect the double pulse, rather than absolute numbers.
I would consider using two interrupts: a RISING and a FALLING - or even simply a CHANGE followed by detection of what happened - as you need to know both the pulse length and the time between the two pulses, this for the double pulse detection.
This way, if you detect a rising edge, and the time since the last falling is less than say two times the duration of the previous pulse, you instantly know you have the second pulse of the double pulse, regardless of the actual number.
As this is a motor, fastest speed would be something like 10,000 rpm or 700 pulses per second, that's about 1.4 ms between pulses. Definitely requires micros() timing as each pulse would last some 100-200 µs based on that scope image.