Cloth Deformation Prediction Based on Human Motion

In the field of cloth animation, efficient and vivid simulation with deformation details especially small wrinkles has always been challenging. Animation with adaptive meshes is a typical way to improve the quality and efficiency simultaneously. Existing works mostly focus on how to model the adaptive mesh and work out the numerical solution while rarely highlight the problem of predicting deformation trend of cloths. In real as well as virtual scenes, cloth movements are usually driven by human body. Intuitively there is some strong relationship between human motion and cloth deformation. Therefore, it is significant to learn the relationship between them to support the research on mesh resolution adaption. In this paper, we describe and evaluate a new machine learning approach to predict deformation distribution. Firstly, we extract human posture feature from motion data and cloth deformation distribution feature from high resolution cloth animation examples. Then we build and train a neural network to predict deformation distribution on cloth based on human motion. The predicted result has high reliability and could be used in further dynamic adaption of cloth mesh in animation.