Depth Image Processing Technique for Representing Human Actors in 3DTV using Single Depth Camera

In this paper, we propose depth image processing technique for 3DTV. A depth camera can simultaneously capture color and geometry information of any objects at video frame rate. However, it cannot acquire some regions whose surfaces are shiny and dark. Therefore, to represent natural human actors, the lost hair region should be recovered. The recovery consists of three steps: lost region detection, boundary depth estimation, and hair shape estimation. Another problem of original depth image is the presence of too much noise to provide haptic interaction. Thus, depth image is smoothened after performing a series of operations. As a result, we can provide an accurate force feedback for end-users. The proposed method automatically generates refined and smooth depth image stream. Finally, we can render 3D scenes using the triangulation of depth images and texturing with synthesized background and CG object. The rendered scenes can be used for 3DTV with haptic interaction.