Abstract: Limb motion analysis and recognition is critical for somatosensory interactions.In this paper,a new posture sequence finite-state machine(FSM) method is proposed for improving the efficiency and versatility of natural motion recognition.Firstly,a user-centric limb joint coordinate system is setup to transform the skeletal data of the user motions from the device space to the user space,and a three-dimensional grid model is built to eliminate the users' individual differences.Secondly,the joint motion sequences of predefined motions are sampled and analyzed using the limb joint feature vectors,which are defined in the limb joint coordinate system,with spatial motion vectors,motion time intervals,and spatial distances of joints being employed to depict the limb motion features.Finally,the limb motion trajectory is represented by the joint movement regular expressions,and the posture sequence FSM is constructed to recognize the predefined motions online.Experimental results on 17 kinds of predefined motions show that the proposed method has the merits of high recognition rate,good scalability and versatility,and is suitable for somatosensory interaction applications.