Real-time Full View 3D Human Reconstruction using Multiple Depth Scanners

Abstract

We present real-time solution for 3D human body reconstruction with multiple RGB-D cameras. Our system uses four consumer RGB-D cameras, each is located at approximately 90° around the freely moving human body. A single mesh is constructed from the captured point clouds by removing the estimated overlapping region iteratively from the boundary. We develop the cell based mesh construction algorithm resilient to various conditions by considering the direction of the camera and of the mesh boundary. The algorithm enables the holes or occluded regions have more chance to be recovered by another view and feasible to the parallel processing. Finally, the calibrated RGB data are merged with the constructed mesh to be viewed from an arbitrary direction. Most of the algorithms are implemented with GPU pipeline for the real-time processing.