Modeling of fabric motion based on small videos

Abstract

Fabric recognition remains an ongoing challenge caused by complicated nonlinear and anisotropic elastic behavior of textiles. In this paper, we devise a computer vision system to automatically analyze the interaction forces of fabrics from videos. We start developing this textile force model by considering particle advection based on optical flow. Our model, which utilizes optical-flow measures of particle interactions within fabrics, incorporates the latent dirichlet allocation model to furnish textile classifications. We validate the correctness of this model by classifying fabrics from videos into an array of textile categories. Our empirical study demonstrates that the fabric classification system powered by our force model is capable of distinguishing different textile materials recorded in videos. We also demonstrate that the interaction force of the same kind of textiles is equal under various unknown wind forces from videos. Our approach to modeling interaction force of textile materials opens a new door to research on future textile recognition technologies.