PROBLEM TO BE SOLVED: To easily and accurately identify the gripping state of a terminal without requesting any special action to a user, or mounting any special device on the terminal.SOLUTION: In an operation recognition system 1, the probability density function of a time variation state is calculated by using gripping data from the gripping start to end of a terminal, and a feature vector is acquired by using dissimilarity between the acquired probability density function and a reference probability density function as a featured value, and the learning of a feature vector for personal collation based on similarity between the feature vector and the representative feature vector of each gripping action corpus prepared in advance is performed, and a similarity index between a feature vector for inspection acquired for gripping data for inspection in the same procedure and the feature vector for personal inspection is calculated, and the gripping state of the terminal is determined by using the acquired similarity index.COPYRIGHT: (C)2015,JPO&INPIT【課題】ユーザに特別なアクションを要求せず、端末に特別な装置を搭載しないで、簡易に且つ精度良く端末の把持状態を識別する。【解決手段】動作認識システム1では、端末の把持開始から終了までの把持データを用いて時間変動状態の確率密度関数が算出され、得られた確率密度関数と基準の確率密度関数との相違度を特徴量として特徴ベクトルが求められ、該特徴ベクトルと事前に用意した各把持行動コーパスの代表特徴ベクトルとの類似度に基づく個人照合用特徴ベクトルの学習が行われる。また、同様の手順で検証用把持データに対し求められた検証用特徴ベクトルと個人照合用特徴ベクトルとの類似尺度が算出され、得られた類似尺度を用いて端末の把持状態が判定される。【選択図】図1
