ABSTRACT
ABSTRACT: Recently, surface Electromyography (sEMG) signals have been widely used for the detection of limb actions and applied as control signals in a Human-Computer Interaction (HCI) system. As previous studies indicated, typical natural human actions are often composed of several sequential movements, whereas different human actions are composed of different sequential movements. These observations could be addressed and taken advantage of to improve the performance of the HCI system, if the sequential movements could somehow be efficiently detected and correctly recognized. In this paper, an approach to recognize human action by detecting the sequential movements from the sEMG signals of upper limb muscles, was proposed and realized. The sEMG signals of Anterior Deltoid muscle (AD), Biceps Brachii muscle (BB) and Flexor Digitorum Superficialis muscle (FDS) were acquired, and the features of Mean Absolute Value (MAV) and Waveform Length (WL) were then extracted. Seven-class actions were recognized by means of decision-tree, and the classification accuracy of MAV and WL reached to 95.24% and 94.05%, respectively. In addition, the features of MAV and WL were fused on both feature-level and decision-level, and the classification accuracy increased to 96.43% and 98.81%, respectively. The results indicated that the sequential feature of sEMG signals could be exploited effectively and multi-feature fusion method could increase the classification accuracy further. The approach of this paper contributed to realize a more practical HCI system by recognizing human upper limb movements sequentially.
