Dynamic 3D Models Generated From Multiple Video
Kensuke HISATOMI, Kimihiro TOMIYAMA, Miwa KATAYAMA and Yuichi IWADATE↓summary
- summary
- We propose a method of generating dynamic 3D models for each frame of video captured by multiple cameras surrounding a human subject and a method of mapping textures that gives "the feel" of a material by mapping camera images to the surface of the human model. In generating dynamic 3D models, we adapted the stereo matching method supported by the regional surface features of the visual hull to reconstruct accurate models stably. For the texture mapping, we adapted a method to blend three camera images that are closest to the viewpoint and visible from a target polygon and to map the blended texture to the surface of the generated models to improve arbitrary-viewpoint images.