Improving Free-Viewpoint Video Content Production Using RGB-Camera-Based Skeletal Tracking

Free-Viewpoint Video (FVV) is a type of volumetric content in which an animated, video-textured 3D mesh of a character performance is constructed using data from an array of cameras. Previous work has demonstrated excellent results when creating motion graphs from FVV content, but these techniques are often prohibitively expensive in practice. We propose the use of skeletons to identify cut points between FVV clips, allowing a minimal set of frames to be processed into a 3D mesh. While our method performed with 2.8% poorer accuracy than the state-of-the-art for our synthetic dataset, cost and processing time requirements are dramatically reduced.