Erik is quite right. Signal is changing abnormally. Also 'double beat' detection is very complicated. You can see these difficulties when you test such stuff built for sportsmen. Sometimes pulse goes well beyond 250 beats per minute, next period only 110 beats are displayed.
There are many devices on the market which show completely nonsense readings even when person is not in any motion.
When I studied physics I was in contact with a professor working about this topic and I know that he must use a DSP and lots of mathematics. I'm not quite sure but I believe that he did something like sliding FFT and sophisticated averaging.
All solutions with Schmitt-triggers and fixed threshold will only satisfy in the most trivial situations, if at all.

Kai