Real Time Motion Authoring of a 3D Avatar

A diverse number of avatars that express human motions are produced in the fields of film, computer games, and virtual reality. In order to simulate human motions realistically, data are collected through a sensor-based method such as motion capture and are organized into a database; motions are then generated on the basis of this database. Human body motions can be reproduced naturally through a vast amount of data, but complicated calculations are necessary to generate these motions, and thus, it is not easy to generate an interactive 3 dimensional (3D) avatar in real time. In order to generate the motions of a 3D mobile avatar in real time, in this study, we implement a dataset with an intuitive classification through pattern recognition and propose the mapping of various avatar motions using a flexible and adaptive hierarchical algorithm. For real time motion authoring, we use a support vector machine (SVM) to classify the types of motion pattern data input from magnetic, angular rate, gravity (MARG) sensor.