Dealing with Variability When Recognizing User's Performance in Natural Gesture Interfaces

Recognition of natural gestures is a key issue in videogames and other immersive applications. Whatever the motion capture device, the key problem is to recognize a motion that could be performed by different users at interactive time. Hidden Markov Models (HMM) are commonly used to recognize the performance of a user but they rely on a motion representation that strongly affects the global performance of the system. In this paper, we demonsrate that using a compact motion representation based on Morphology-Independent features offers better performance compared to classical motion representations especially for users whose data were not used for training.