Real-Time 3D Model Reconstruction and Mapping for Fashion

Recent developments in the apparel market created new tasks for computer vision application. One such challenge is designing a system for trying on clothes virtually in real-time. Proposed solutions require setting up a special stationary system for body capture and mostly generate animated avatar instead of overlaying the garments in real-time. We propose a framework that utilizes state-of-the-art monocular-based 3D skeleton reconstruction and parametric body generation techniques allowing to operate under constrained resources, such as smartphones. In addition, we also consider the problem of dealing with visual artifacts as a result of 3D projection on a real-time image and design a solution to reduce them based on iterative closest point method.