Aiming at problem that in ballistocardiogram（BCG） signal heartbeat waveform has characteristics of no certain model and is influenced by various factors

propose a heartbeat recognition algorithm based on self-learning.During self-learning process

a group of curves with high correlation which are extracted from the BCG signals by means of clustering analysis

and make it as heartbeat model.The heartbeat model is matched with BCG signals

so catch heartbeat signal

so as to obtain heartbeat cycle.By experimental verification

error of heartbeat cycle which is calculated based on this algorithm is within ±2 %

besides

the algorithm is proved to be accurate and practical on Android platform.