HMM with improved feature extraction-based feature parameters for identity recognition of gesture command operators by using a sensed Kinect-data stream

Artificial internet of things technology encourages the development of robots and sensors. Kinect sensors with excellent human gesture recognition and robots with smart interactions with people are expected to have numerous innovative applications. In robotic sport instructor systems for rehabilitation and exercise training, identity recognition of the gesture operator is a crucial problem. With operator identity recognition, the gesture classification model owned by the operator can be further adjusted using the operator's active gestures, and a user-adaptive sport instructor robot system can then be achieved. This study developed an identity recognition method to classify gesture operators in which an improved feature extraction scheme that considered the practical height of a person was introduced for effective identification. According to the improved feature extraction design, a 40-dimension feature vector with the physical characteristics of the human skeleton was further developed as the feature parameter for enhancing identity recognition. A hidden Markov model (HMM) with fine considerations of continuous-time gesture variations was adopted as the recognition model for identifying ten gesture operators in the sport instructor robot system. Experimental results demonstrated the superiority of the presented approach because the constructed corresponding ten HMM active user models with the improved feature extraction-based feature parameter exhibited an excellent average identity recognition accuracy of 87.67% among all ten players, with each of them making six specified sport rehabilitation actions.