A kinect-based gesture command control method for human action imitations of humanoid robots

This paper develops a Kinect-based gesture command control method for driving the humanoid robot to learn human actions. The popular Kinect sensor is well-known for its high performances on gesture recognition. This work uses three different recognition mechanisms, dynamic time wrapping (DTW), hidden Markov model (HMM) and principal component analysis (PCA)-based eigenspace approaches for performing the recognition task of the specified human active gestures captured by the Kinect sensor. The recognized gesture command is then used to control the action of a humanoid robot where the humanoid robot will imitate the human active gesture according the content of the gesture command. By the presented method, the humanoid robot will effectively learn the human action. A series of experiments on gesture recognition and humanoid robot control are done to compare the gesture command recognition performance of three recognition methods and evaluate the similarity degree of humanoid robot imitations on the gesture command-indicated human active gesture.