Partial or complete loss of the upper limb motor function has great impact on the activities of daily life (ADL) of post-stroke survivors. To improve the rehabilitation effect of fine motor function of forearms, a couple of recent studies focused on methods that try to decode the limb motion intent of patients through physical exercises. However, there exist a few studies on real-time active rehabilitation method for the classification of multiple hand movements. In the current investigate, a pattern-recognition based rehabilitation environment was set up using high-density surface electromyogram (HD-sEMG) and the real-time classification performance of 21 forearm motions was investigated with eight healthy subjects. The results showed that the average motion completion rate across all subjects was 91.17%+/- 2.86%, which suggests the potential of intention-initiated approach in assistive rehabilitation technique.