Behavior Classification and Description Generation on 2D Human Postures in Care Facilities
Wataru Takano, Haeyeon Lee · The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec) · 2019
This paper describes a novel approach to classifying whole body motions from estimated human postures in 2D camera images, and subsequent generating their relevant descriptions. The motions are encoded into the stochastic motion models that are referred to as motion primitives. Words are connected to the motion primitives, and word n-grams are represented in the stochastic manners. More specifically, the motion observation is classified into the motion primitive, from which its relevant words are generated. These words are arranged in a grammatically correct order to make the descriptions for the observation. This approach was tested on actions performed by the elderlies in the care facility and its validity was demonstrated.