A Methodology to Create 3D Body Models in Motion

Abstract

Size, shape and posture are fundamental features of digital human models (DHM) to obtain accurate virtual simulations of the ergonomics of products and environments. Research on 3D body scanning, processing and modelling have enabled the generation of avatars representing specific populations and morphotypes in standing and seated postures being the basis to define size and shape of DHM. Posture is implemented with biomechanical models of the human movement. Most of the research is focused on posture control and movement tracking to analyze the variability in different contexts (e.g. driving, performing a working task). Motion capture technology used for this purpose, requires a limited number of sensors or reflective markers attached to the body according to the definition of body segments. 3D body scanning and motion capture are both technologies currently used to analyze human body shape and biomechanics to apply it to enhance digital human models. These technologies may converge on the so-called temporal 3D scanners or 4D scanners, a new technology recently developed to scan the body in motion. With this technology, it is possible to obtain sequences of dense 3D point clouds representing the movement of the body. In this paper, a novel methodology to create realistic 3D body models in motion is proposed. This method is supported by a new 4D scanning system (Move 4D) and a data driven-model. Move4D is a modular photogrammetry-based 4D scanning system. It consists of a set of 12 synchronized modules to scan full bodies with texture in motion. It can capture up to 180 fps with a resolution of 2 mm. The algorithms have been conceived and optimized to automatically process the series of raw point clouds captured. They rely on a data-driven body model including shape, pose and soft-tissue deformation trained with a large database and a deep learning model. The process is fully automatic and does not require any interactive landmarking or revision. The 3D outcome of this methodology is one noise-and artefact-free watertight mesh per frame and a model of shape, pose and soft-tissue that can be rigged with a 23-joint skeleton. This type of outcome permits their use for many applications such as simulations, augmented and virtual reality (AR/VR) or biomechanical analysis purposes.