Abstract
For the management of a large number of motion data for humanoid virtual actors we propose to use a database system to store and retrieve motion data with additional information. We show key questions lying behind the different classification schemes for human motion. Having obtained these criteria, we deduce the requirements a database has to fulfil for coping different classification schemes. We describe the AMOBA database system which allows to store and retrieve human movement data with an expandable set of classification schemes. To implement classification schemes we annotate movements with specific characteristics.