Abstract
This paper describes a new approach to daily human activities recognition: 1) decouple the problem of object recognition and that of behavior recognition, 2) resolve the former by 3-D tagging/sensorizing technology, and 3) develop a model for recognizing human behavior from such a sensorized objects. This paper proposes “an object sensorizing system” which realizes a function for robustly observing and recognizing daily human activities using real-object-shaped sensors. The object sensorizing system consists of three layered components. 1) Physical sensor layer: an ultrasonic 3D tag system outputs ID and 3D positions of objects. This function contributes to avoiding a problem of feature-based object recognition. 2) Virtual sensor layer: physical characters of real objects are abstracted through simplifying their 3D shape and unique functions using prepared virtual sensors. 3) Object sensor layer: by defining behavior to be recognized using the abstracted objects, this layer enables to create object-shaped sensors which are added a new function for outputting recognized behavior while maintaining their original functions. This paper also reports an experimental system for demonstrating the proposed method.
