GENERATION OF OPEN-DOOR-ACTION ACCORDING TO HIP TRAJECTORY USING CONTINUOUS HMM TECHNIQUE(International Workshop on Advanced Image Technology 2006)

概要

論文の詳細を見る
Opening a door and going through it is a trivial action in ordinary life. For a virtual agent, however, actions like approaching the doorknob from an arbitrary position, pulling it, and following the door movement, all have to be taken into account in order to generate the computer graphics animation for that action. In this paper, we make use of machine learning techniques which can build models for the animation synthesis of opening a door. First we use motion capture system to acquire motion data for the typical action. The data set along with restraint conditions are fed into continuous Hidden Markov Models (HMMs) which can learn the action of opening a door. Restraint conditions are what should be considered while moving, such that the foot on the floor must not slip. To generate a new animation, the hip trajectory of a character is applied to the model, then the highest probability of state transitions in HMMs is calculated and the corresponding sequence to the transitions is generated automatically. As a result, we generate a door opening action which consists of parallel and serial motions.
社団法人電子情報通信学会の論文
2006-01-02