Retrieval of Motion Data Using Posture Images on Self-Organizing Maps

Abstract

This paper proposes a method for retrieving motion clips using key postures. This method generates a self-organizing map (SOM) from motion files to be retrieved, and then extracts desirable frame segments by specifying postures corresponding each node. Intuitive interfaces are constructed by supplying icons of representative postures that are determined by using a clustering of postures distributed on the map.