Deformable Modeling for Human Body Acquired from Depth Sensors

TL;DR

This paper introduces a method to reconstruct and animate complete 3D human body models over time using data from a single depth sensor, combining surface assembly, mesh warping, and volumetric techniques.

Contribution

It presents a novel integrated approach for deformable 3D human body modeling from single depth camera data, including surface assembly, mesh warping, and volumetric fusion.

Findings

Successfully reconstructs complete 3D models from partial scans

Effectively aligns and smooths models over time

Handles missing data and errors in depth captures

Abstract

This paper presents a novel approach to reconstruct complete 3D deformable models over time by a single depth camera. These are the steps employed for deforming objects from single depth camera. The partial surfaces reconstructed from various times of capture are assembled together to form a complete 3D surface. A mesh warping algorithm is used to align different partial surfaces based on linear mesh deformation. A volumetric method is then applied to combine partial surfaces, fix missing holes and smooth alignment errors.