A Lightweight Low-Cost Multi-View Capture System for Dynamic Human Reconstruction
Kai Altwicker, Dennis Amuser, Arthur Kehrwald and Arnulph Fuhrmann
In: Virtuelle und Erweiterte Realität – 23. Workshop der GI-Fachgruppe VR/AR, 2026
Abstract
High-quality avatar reconstruction often relies on dense, temporally consistent multi-view observations, yet acquiring suitable training data typically depends on expensive professional volumetric studios. This paper presents the VolumanXR Volumetric Capture System (VCS), a lightweight and low-cost multi-view capture platform for affordable avatar acquisition using commodity hardware. The realized system employs 68 Raspberry Pi 4B units with Camera Module 3 sensors arranged in a modular octagonal rig and combines centralized control, shared-time Wi-Fi synchronization, frame-drop awareness, and automated per-camera focus calibration. These design choices address the practical challenges of producing reconstruction-ready datasets for modern avatar generation methods, including Gaussian-splatting-based approaches. The system is evaluated through synthetic and practical experiments covering reconstruction quality, synchronization, and focus calibration. Results indicate that commodity hardware can provide frame-aligned, reconstruction-suitable multi-view data for Gaussian-splatting-based human reconstruction within the evaluated operating conditions, offering a comparatively accessible alternative to professional volumetric capture systems.
Paper
Preprint:
