Integration of pose recognition for a person wearing short or long sleeves
Takuya Kawano, Kazuhiko Yamamoto, Kunihito Kato, H. Hongo · 2003
In this paper we propose a TV control system for aged and bedridden people. We have been constructing the interface, which changes TV channels and operates volume controls using pose recognition. However, users were required to wear a short-sleeved shirt for recognition of poses using skin color regions as arms. To allow the wearing of long sleeves, we improve our system to introduce the relation of the position of higher local auto correlation features to identify the pose. First, face and hand regions of the user are detected using skin color from the input image. Since it assumed that the width of a face region should be larger than arms, the face region is decided from the detected skin color regions using an erosion operation. After detecting the face region, a pose recognition area that is decided on the basis of face position is divided into 9 (3 /spl times/ 3) regions. Features are extracted from the 6 areas because the arms appear within the 6 areas. Using our algorithm, our system can recognize poses even if the user wears a long-sleeved or short-sleeved shirt.