@techreport{oai:ipsj.ixsq.nii.ac.jp:00226135,
 author = {小栗, 芙美果 and 伊藤, 弘大 and 前田, 竜矢 and 田中, 久範 and 伊藤, 雄一 and Fumika, Oguri and Kodai, Ito and Tatsuya, Maeda and Hisanori, Tanaka and Yuichi, Itoh},
 issue = {29},
 month = {May},
 note = {In this study, we propose an input device using a soft and transparent gel and a system to estimate 3D interactions to the soft materials using the device. The device consists of a soft and transparent urethane resin, infrared LEDs, and infrared phototransistors. By deforming the device, the LED light intensity passing through the transparent gel changes. We propose a system to esitimate the deformation of the device by machine learning from the feature values of phototransistors that changes in response to the light. As an evaluation experiment, we classified 13 different interactions using two hands and evaluated the accuracy of the system. We acquired data 10 times for each interaction and classified them using SVM. The classification accuracy was evaluated by stratified 10-part cross-validation, and it was confirmed that the interaction could be classified with 98.5% accuracy. Then we also estimated the bending curvature of the device. We bent the device along the five different radius cylinders and acquired 10 data for each. The curvature is calculated from the radius of cylinders. We evaluated the accuracy of SVR estimation by 10-part cross-validation and got 0.93 coefficient of determination(R2)., In this study, we propose an input device using a soft and transparent gel and a system to estimate 3D interactions to the soft materials using the device. The device consists of a soft and transparent urethane resin, infrared LEDs, and infrared phototransistors. By deforming the device, the LED light intensity passing through the transparent gel changes. We propose a system to esitimate the deformation of the device by machine learning from the feature values of phototransistors that changes in response to the light. As an evaluation experiment, we classified 13 different interactions using two hands and evaluated the accuracy of the system. We acquired data 10 times for each interaction and classified them using SVM. The classification accuracy was evaluated by stratified 10-part cross-validation, and it was confirmed that the interaction could be classified with 98.5% accuracy. Then we also estimated the bending curvature of the device. We bent the device along the five different radius cylinders and acquired 10 data for each. The curvature is calculated from the radius of cylinders. We evaluated the accuracy of SVR estimation by 10-part cross-validation and got 0.93 coefficient of determination(R2).},
 title = {透明なゲルを用いた柔らかい入力インタフェースに関する検討},
 year = {2023}
}