关键词: 三维虚拟试穿, 人体动态特征, 协同跟踪, 服装骨骼点分层模型, 整体-局部分解

Abstract:

Aiming at the dynamic perception of personalized postures and the tracking and matching of human body's posture by clothing modeling, this paper established the hierarchical 3-D clothing skeleton points model based on two types of the motion features of the human postures, distinguishing the trunk (global) from the limbs (local), and integrating the clothing model with the body circumferencial features extracted using Kinect camera. In parallel, the global-local singular value decomposition algorithm (GL-SVD) was designed to hierarchically control the deformation of 3-D clothing skeleton model cooperatively. The results from virtual try-on test system show that relative tracking error of the new method is reduced by about 10%, the real-time tracking accuracy and matching precision between the clothing model and the human body is obviously improved. This new technique supports the complex posture (such as leg lifts, lunges) trying-on scenarios of various clothing types such as skirts, pants, tops, and more.

Key words: 3-D virtual try-on, dynamic features of human body, cooperative tracking, hierarchical model of clothing skeleton points, global-local singular value decomposition