Towards accessible content creation of real world objects for virtual environments
3D reconstruction is the general problem of creating 3D models from real world objects. In today's movie and games industry,
there is an increasing demand for using real world content as assets in production. In general, however, 3D reconstruction is a
challenging problem, and current techniques only allow for production-ready results given a combination of expensive equipment and
This thesis is a collection of three papers that address various aspects of this general problem of 3D reconstruction,
with the aim of lowering the bar for making usable real world content.
In Paper I, we address the problem of storing and streaming time varying geometry for e.g.\ free-viewpoint video, which
otherwise has too high bandwidth requirements to be streamed efficiently. We use a memory-efficient structure based on compressed
voxels to store the data, in which we can send only incremental updates to the geometry in each frame.
In Paper II, we implement an end-to-end real-time pipeline for free-viewpoint video communication.
The pipeline uses a set of ordinary webcams as input and do all processing on a single desktop computer. Even with these
limitations, we show that we can produce free-viewpoint video with agreeable quality in real-time.
Paper III addresses the problem of accessible and accurate modeling of static real-world objects.
Given a set of calibrated input images, we have developed an interactive tool that makes 3D reconstruction with multi-view stereo more
accessible. This interactive reconstruction has several advantages over automatic 3D scanning, since we obtain correct topology by design
as well as information about visibility and foreground segmentation.