An Effective One-shot Body Part Multi-View Reconstruction Device with Self-calibration Capabilities
Keywords: Auto-Calibration, Multi-View Stereo, Active Stereo, Point Cloud Processing
Abstract. This paper introduces a custom-built low-cost camera ring device designed for automatic cast synthesis, able to accurately and instantly scan body parts. The scanned mesh will be used as a backbone model for the cast design and 3D printing. The system is based on the multi-view active stereo principle and it is composed of a circular array of 16 synchronized cameras (Fig. 1) and 4 equally distributed IR pseudo-random laser pattern projectors. We employ a custom multi-view stereo reconstruction pipeline based on (Schönberger et al., 2016), which guarantees optimal results without the downsides of the supervised data-driven multi-view stereo algorithms, i.e. data collection and ground truth labeling. Additionally, inspired by (Duda and Frese, 2018), we propose a novel, automated calibration system to extract intrinsic and extrinsic camera parameters which are required to perform robust multi-view stereo reconstructions.