The non-contact life parameters detecting systems can detect many important life parameters such as heartbeat and breathing at certain distance. In practice, the self-dithering interference of the system is produced by some factors; it makes the signal-to-noise ratio (SNR) of the recorded signal lower. This paper is discussed an adaptive technology based on a variable step-size LMS algorithm, which is used to eliminate the self-dithering interference of the system and hence to raise the SNR of the detecting signal. A method to obtain the reference signal by vibration sensor is brought forward and the optimum parameters settings are discussed. The results show that the reference signal is an integral part of output signal of the vibration sensor and has a relationship with the setting direction of the sensor. The conclusion is that the method is effective and feasible; also the algorithm is well convergent.