@techreport{oai:ipsj.ixsq.nii.ac.jp:00186645,
 author = {風間, 康介 and 堀内, 靖雄 and 下元, 正義 and 黒岩, 眞吾 and Kousuke, Kazama and Yasuo, Horiuchi and Masayoshi, Shimomoto and Shingo, Kuroiwa},
 issue = {16},
 month = {Mar},
 note = {本報告では，言語情報を用いた Kinect による連続指文字認識手法について検討する．一般的に，指文字は単語として連続で呈示されるため，指文字を連続で認識できる必要がある．我々の先行研究では，HMM によって腕の動作の認識を行い，腕の動作ごとに学習した SVM によって指文字を認識する手法を提案したが，HMM の認識誤りの際に認識に失敗してしまうという問題点があった．そこで我々は，動作認識誤りの際にも指文字の認識が行えるように動作の認識と手型の認識を分離して行い，単語辞書に基づく言語情報によって指文字単語を認識する手法を提案した．評価実験を行ったところ，認識対象語彙を制限した場合で，話者クローズでは 98.9%，話者オープンでは 93.5% の認識率を得た．, In this paper, we will discuss a finger spelling recognition based on linguistic information. Generally, a finger spelling is performed successively as a word. Therefore, it is necessary to recognize the finger spelling continuously. We have previously proposed continuous recognition of finger spelling. This method to recognize movement of arm by HMM, and recognize finger spelling by SVM which trained finger spelling for corresponding movement of arm. However, if it fails to recognize movement of arm by HMM, the finger spelling recognition fails. Therefore, we proposed a method to recognize a finger spelling word using linguistic information based on a word dictionary by independently recognition of movement of arm and recognition of hand shape, that can recognize correct finger spelling, even if it fails to recognize movement of arm by HMM. As a result of the recognition experiment, we obtained recognition rate of 98.9% for speaker closed condition and 93.5% for speaker open condition when recognition word vocabulary was restricted.},
 title = {言語情報を用いたKinectによる連続指文字認識手法の検討},
 year = {2018}
}