CNNによる爪色三軸触覚センサの高精度化

抄録

In our previous study, we developed a fingernail color sensor capable of identifying the magnitude and direction of the force acted on fingers. The prototype is equipped with a small CMOS camera and green LEDs, and collected a dataset of vertical force and shear force acting on the index finger and fingernail color images using an electronic balance and a load cell. The dataset was trained by CNN to estimate forces from the images. In this study we produced individual CNN models, which learned individual fingernail color change data. RMSEs for nine human subjects were 0.12, 0.12 and 0.29 [N] for F_x, F_y and F_z, respectively. RMSEs of F_x and F_y are significantly smaller than those of previous studies.