Self-Organizing Mapを用いた臥床・寝返り・離床動作の推定

抄録

In this paper, we developed the motion detecting system using the depth image which applied two-dimensional discrete Fourier transform (2D-DFT). The depth images of the user are providing the user’s posture considering the user’s privacy. The depth images applied 2D-DFT shows the spatial frequency. Therefore, if the user's posture is similar even if the user's positions are different on the camera images, the transformed results are almost same. Our system is designed to determine three postures: (1) lying down on the bed, (2) turning over in the bed, and (3) leaving the bed. To determine the three postures, we obtained the number of lines and their angles using Hough transform from the two-dimensional spatial frequency spectrum. And we applied the Self-Organizing Map (SOM) to classify the three motions. In the experiments, we obtained favorable results. Our proposed method could estimate the three motions with blanket especially turning over in the bed and leaving bed motion.