You are right, it's surprising the great work of the device when there are more than one present.
Problems appear when you have two subjects A and B a bit close, in that case if Device A lost the hand of its subject it will take the hand of subject B if its into his detection box.
I think about the device as a camera, if I reach it to se the hands, whatever the wavelength and filters I use, it will track it right.
Maybe someone has some technical information about the cameras ¿?