Kinect驱动的实时虚拟人物试衣  

Real-time Virtual Model Fitting Moved by Kinect

作　　者：唐勇[1,2] 李梦琪[1,2] 刘宇涵[1,2] 吕梦雅[1,2] 

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004 [2]河北省计算机虚拟技术与系统集成重点实验室,河北秦皇岛066004

出　　处：《小型微型计算机系统》2016年第12期2765-2768,共4页Journal of Chinese Computer Systems

基　　金：河北省自然科学基金项目(F2012203084)资助

摘　　要：虚拟试衣是虚拟现实技术研究的热点,实时2D贴图法不足以展现布料的弹性、褶皱等属性;静态三维模型法建模精细但非实时,因此,虚拟试衣的真实感和实时性成为了一对突出的矛盾.首先,采用添加约束力的质点弹簧模型建立布料模型,通过约束方程实现布料的拉伸、碰撞等自身属性,避免了隐式积分的耗时问题,增强布料真实性的同时大大提高了布料建模的速度;其次,用骨骼变换矩阵实现虚拟人物运动,解决了静态模型不能实时展现运动布料的问题;再次,用球面蒙皮法对虚拟人物蒙皮,避免了由于过渡拟合出现的皮肤变形现象,提高了虚拟人物真实感;最后,用邻域分割R值化方法提取人体轮廓深度信息,减少了时间复杂度,并对轮廓边缘进行双边滤波处理,提高了Kinect识别人体的准确率.多组实验结果表明,本文实现了Kinect驱动的实时虚拟人物试衣,并增强了虚拟试衣真实感.The virtual model fitting is the hot spot of the research on virtual reality technology. However, the real-time 2D mapping method is inadequate to show the fabric＇ s elasticity,folds etc. The static 3D model method is fine but unreal-time. So,the reality and the real-time have become a pair of contradiction during virtual model fitting. Firstly, the constraint spring model was used to build the cloth model. We achieved the cloth＇ s drawing and the collision by the constraint equation. The implicit integral time consuming prob- lem was avoided. The reality and the real-time were improved at the same time. Secondly, bone transformation matrices were used to solve the problem that the static model can ＇ t show the cloth ＇ s moving in real-time. Moreover, in order to make the virtual human body be more real, the spherical skin was applied to avoid the skin＇ s deformation due to transition fitting. Finally, the R value on the domain decomposition was used to obtain the body outline, and then to reduce the time complexity. Thanks to the bilateral filtering for outline＇ s edges, the accuracy in Kinect＇ s recognition was improved notably. Several experiments were done. And the results show that we realized the real virtual model fitting driven by the Kinect in the real-time and improve the realistic during the virtual model fitting.

关 键 词：虚拟人物试衣 约束力弹簧 球面蒙皮 邻域分割R值化 KINECT 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]