Abstract:A new method of binary decision analysis based GM threshold was proposed, applying to the research on recognition of human upper limb motion. The human motion acceleration acquisition system was designed, based on selection of the MEMS triaxial acceleration sensor ADXL345. Through defined types of actions to do experiments, after obtained the three dimensional (3-D) acceleration signal of the body to perform upper limb actions, first the wavelet threshold denoising was preprocessed, extracting the Geometric Mean (GM) of upper limb actions as feature to select threshold. The experimental results show that, combined with wavelet analysis theory, this method based on GM threshold achieves the recognition of the type of action of the human upper limb, and validates the simplicity and efficiency of the system design and the theoretical algorithms, which has some reference value to research on human motion recognition based on acceleration sensor.