Machine Learning-driven Classification of Hand Motion for the 3D-proximity-sensors Unit

抄録

This paper proposes a machine learning-based method to identify human hand motion using a 3D capacitive proximity sensor based on multiple sensing electrodes, which was developed in our previous studies. Although the sensor can detect nearby objects, determining their position and motion directly from the nonlinear outputs of the sensor is difficult. This study proposes a random forest method to identify the direction of movement of a human hand passing above the 3D proximity sensor unit. The time-series data obtained by combining the outputs of three channels are classified into four directions: upward, downward, rightward, and leftward. Experimental evaluation reveals that the proposed method achieves over 95% classification accuracy in all four directions.