無線型加速度センサと畳み込みニューラルネットワークを用いた触対象の分類

抄録

Many methods of collecting haptic information were proposed by many reserchers. In our previous research, we proposed a method of collecting accelerations in haptic behaviors using wireless accelerometers. The method enables to collect haptic information easier than other previous methods. However, classification of the information was not researched enough. Therefore, in this paper, we collected haptic information using the method and classify the information using Convolutional Neural Network. We performed an experiment to classify rubbed objects. Our system classified 90 kinds of haptic information by 88.7%. In addition, we performed an experiment to classify rubbed objects in real-time. Our system classified 15 types of object, 71.0% on average.