Volumetric Video Tutorial

May 3, 2021·
Nikolaos Zioulis
Nikolaos Zioulis
· 1 min read
Volumetric Capture @ VCL3D
Abstract
While high-end setups currently support different facets of volumetric capture technology applications like content creation and live telepresence, there is a need to transition towards lower cost, portable setups for digitizing human performances, which are alsomore suitable for experimentation and accessible research. To encourage progress towards this, a low-cost Volumetric Capture system was developed and made openly available along with documentation covering both its software and hardware aspects.
Date
May 3, 2021 4:30 PM
Event
Location

Online

In this tutorial, apart from presenting and documenting its operational details, we will offer insights for its various design choices, the challenges that were encountered, and the lessons learned. These span both hardware and software topics like the selection of sensors, the balancing of processing power and costs, the design of the system, and even tripod mounts to ensure the portability of the setup. The presented Volumetric Capture system has been remotely and on-site deployed in various places around the globe as one of its central design goals apart from affordability was its portability and usability. The tutorial will also describe how these two important components for the aforementioned aspects have been addressed, namely the spatial (i.e. StructureNet) and temporal alignment of the sensors, in ways that are sensor-agnostic – allowing for hybrid-sensor deployments – and scalable, offering flexible capturing scenarios. However, as expected, these design choices come at a cost, and the limitations of this approach will also be presented and discussed. Finally, the tutorial will present the work that has been supported and enabled by this low-cost volumetric capture system, showcasing potential uses. More importantly, taking into account the ongoing data-driven revolution, such systems can be used for easily collecting volumetric datasets, with an example being HUMAN4D.