Virtual Human Bodies with Clothing and Hair: From Images to Animation

of “Virtual Human Bodies with Clothing and Hair: From Images to Animation” by Peng Guan, Ph.D, Brown University, May 2013 Realistic clothing and hair animation are necessary for many applications such as special effects, gaming, and on-line fashion. Thanks to the advances in computer graphics, highly realistic clothing and hair animation is common in recent animated movies by use of Physics Based Simulation (PBS). Clothing and hair modeling/animation are also important to other fields such as computer vision. However, PBS methods create clothing/hair that are specific to a particular body model and there is no good way to invert the PBS process to fit the parameters of the generative model to images. Furthermore, PBS for clothing requires a significant amount of manual work to find the right size for every instance of a body and prepare the clothing for simulation. This makes PBS unsuitable for applications that involve various body shapes, such as virtual fashion and crowd simulation. We believe a new generative clothing model will address these problems. In this thesis, we describe a complete process from estimating 3D human body form using image evidence to animating the body with data-driven, low-dimensional 3D clothing and hair models. First, we describe a solution to estimating 3D human body shape from a single photograph or painting using multiple image cues including silhouettes, edges, and smooth shading. Second, we explore a 2D clothing model in which the clothing is modeled as a closed contour of points that are offsets from corresponding body contour points. We show the increased accuracy of 2D body shape estimation and clothing type classification using such a model. Third, we focus on modeling the appearance of the 3D body and propose a complete system for simulating realistic clothing on 3D bodies of any shape and pose without manual intervention. Fourth, we also present a 3D hair model that performs hair animation in real-time, preserves the key dynamic properties of physical simulation, and gives the animator continuous interactive control over hair styles (e.g. length and softness) and external phenomena (e.g. wind). The result is a 3D human body model that can be estimated from images and then animated with realistic clothing, hair, and body movement.