A HMM-based fundamental motion synthesis approach for gesture recognition on a nintendo triaxial accelerometer

In this paper, we show how to use a Nintendo Wiimote triaxial accelerometer as an input device to make a gesture recognition system based on Hidden Markov Model as the kernel recognition algorithm. We adopted a set of basic movements called “Fundamental Motions” as the synthesis units for all the other complex motions. In the preliminary study, we tried to discriminate the digits from ‘0’ to ‘9’. We analyzed this task and found a set of 16 motions are appropriate to be used as HMM modeling units. By using appropriate feature extraction and HMM topology, we can achieve near 98% and 65% accuracy for discrete motions and continuous motions, respectively.