Multi-user Live Video in a Shared Virtual World for Enhanced Group Collaboration

We present a novel method for merging a live video stream of multiple users into a shared virtual world in a large VR display environment. Collaborators at a remote site can be displayed in a virtual world as video textures to support visual communication more efficiently. To generate the video texture of a user in a dynamic background environment such as a CAVE -like system, we use a camera pair which consists of two gray-scale cameras with band-pass filters, and two color cameras. The gray-scale cameras capture infrared images in which the dynamic background on the screen is eliminated, having only the infrared light reflected by the person. This makes it easy to extract the user's silhouette robustly. Using projection histogram we found rough position of the all users' head in the image. We can then find exact corresponding points using cross-correlation between two user silhouette images to determine the 3D position of each user. The extracted user silhouette is warped by rectification and translated to fit into a color image. We can now generate a video texture of the user by masking the background of the user in the color image. Each video texture is then transmitted to a remote site and displayed on a billboard in virtual space.