Development of Wearable Wrist Contour Sensor for Hand Action Recognition

摘要

In this paper, we developed a wrist watch type contour sensor. This sensor is aimed to realize daily life hand action recognition system based on wrist contour shape. The features of this sensor are distance measurement between wrist surface and sensor band by photo reflectors and sensors scan by ring-counter. Contour measurements were conducted about 15 daily tool handling actions using a prototype sensor which is 80mm long. Finally, a three layer neural network model, which was constructed from the experimental data, successfully recognized 77 percent of the data by cross-validation.