Abstract: As a novel technique, compressed sensing, which can reduce energy consumption can promote the development of remote health monitoring systems based on wearable device. Dictionary learning algorithm has attracted much attention because of its improvement of the performance of reconstructing physiological signals in the field of compressed sensing. Usually, conventional dictionary learning algorithms did not consider the implicit correlation inside signals, resulting in that the characteristic of signals cannot be efficiently captured and thus the signal cannot be accurately reconstructed. In this paper, a correlation based dictionary learning algorithm is proposed to apply in compressed sensing, exploit implicit correlation structure inside the physiological signal efficiently, and overcome the shortcoming, poor reconstruction accuracy, of conventional dictionary learning algorithms. Experiments results show that the proposed algorithm can capture the structure of physiological signal adequately, and thus can improve the signal-to-noise ratio for compressed sensing, namely, the compressed physiological signal can be accurately reconstructed.