We introduce a proof-of-concept telepresence system that offers fully dynamic, real-time 3D scene capture and continuous-viewpoint, head-tracked stereo 3D display without requiring the user to wear any tracking or viewing apparatus. We present a complete software and hardware framework for implementing the system, which is based on an array of commodity Microsoft KinectTM color-plus-depth cameras.
Novel contributions include an algorithm for merging data between multiple depth cameras and techniques for automatic color calibration and preserving stereo quality even with low rendering rates. Also presented is a solution to the problem of interference that occurs between Kinect cameras with overlapping views.
Emphasis is placed on a fully GPU-accelerated data processing and rendering pipeline that can apply hole filling, smoothing, data merger, surface generation, and color correction at rates of up 100 million triangles/sec on a single PC and graphics board. Also presented is a Kinect-based markerless tracking system that combines 2D eye recognition with depth information to allow head-tracked stereo views to be rendered for a parallax barrier autostereoscopic display.
Our system is affordable and reproducible, offering the opportunity to easily deliver 3D telepresence beyond the researcher's lab.