View-invariant Hand-posture Recognition Method for Soft Remocon System

View-invariant Hand-posture Recognition Method for Soft Remocon System

  This paper suggests a robust gesture recognition method against the variation of camera viewpoint. The proposed system uses both 2-dimensional appearance features and finger-angles. We use 3 cameras attached to the ceiling. During database-construction, a dataglove is used to capture finger-angles. Due to the variance in the viewing direction, the same appearance feature does not guarantee the same 3-dimensional structure. Likewise, the same 3-dimensional structure does not guarantee the same appearance. The proposed hand-posture database provides the robustness in decision under variations in the viewing-direction. Consequently, it is possible to build a more natural environment. Additional glove-devices are not required in this method.