@article{oai:ipsj.ixsq.nii.ac.jp:00175046,
 author = {長谷川, 達人 and 越野, 亮 and Tatsuhito, Hasegawa and Makoto, Koshino},
 issue = {10},
 journal = {情報処理学会論文誌},
 month = {Oct},
 note = {本研究では，スマートフォン標準搭載のセンサを複合的に用いてDNN（Deep Neural Network）で学習させることで，スマートフォンが利用者の身体上のどの位置に所持されているのかを推定するシステムを開発する．スマートフォンの所持位置が推定できることで，ポケットの中での誤動作防止や，位置に応じた通知方法の自動変更など，様々なコンシューマサポートが実現できる．本研究では，利用者がスマートフォン所持中に最もとりやすい動きである歩行を対象に，ズボン前ポケット，ズボン後ポケット，胸ポケット，内ポケット，ジャケットポケット，鞄，手という所持位置7種類の推定を行う．被験者16人に対して実験を行い，Leave-one-subject-out Cross-Validation（LOSO-CV）で推定精度を評価した結果，81.7%の精度で所持位置7種類を推定し，胸ポケットと内ポケットを区別しない6種類の推定では86.7%の推定精度を達成した．また，センサを複合的に用いることで推定精度が向上するという点や，加速度センサの値の扱い方によって推定精度が向上することを明らかにした．, In this study, we develop a system detecting a smartphone wearing position on the user's body. This system detects the wearing position by DNN (Deep Neural Network) using observed multiple sensor values. If the smartphone wearing position can be detected, it will be applied to some consumer support applications, such as pocket dialing prevention and automatic changing the notification method according to the wearing position. Proposed method detects seven wearing positions such as “in the trousers front pocket”, “in the trousers back pocket”, “in the chest pocket”, “in the inner pocket”, “in the jacket pocket”, “in the hand”, and “in the bag” when the user is walking. We performed an experiment to collect sensor values for 16 participants. As a result of the evaluation by Leave-one-subject-out Cross-Validation (LOSO-CV), proposed method could classify seven positions with 81.7% accuracy. Moreover, in the case that a chest pocket is regarded as the same position with an inner pocket, proposed method could detect six positions with 86.7% accuracy. This paper also describes that using multiple sensors increases the accuracy of detection, and proposed processing method for accelerometer increases the accuracy of detection.},
 pages = {2186--2196},
 title = {深層学習を用いた歩行時におけるスマートフォンの所持位置推定},
 volume = {57},
 year = {2016}
}