虚拟试衣技术是近些年随着人机交互技术发展而大热的研究方向。虽然基于体感的试衣应用已经可以模拟现实中用户的各种动作来观察衣物效果，但虚拟现实环境下的大多数试衣应用仍缺乏衣物与真实人体结合的直观感受而导致实用性不高。增强现实技术则是转变用户购衣消费体验的一个重要的工具。目前，增强现实作为一个较为新颖的方向，AR试衣的相关技术还不成熟，已经实现的应用大多还停留在将二维服装面片融合真实场景的解决方案上，用户得不到直观的衣物与人体之间交互的响应，衣物布料的渲染缺乏物理真实感。

针对这一问题，本文以实现增强现实下的虚拟试衣为目标，提出了一种基于人体点云构造碰撞体，利用碰撞体人体模型来进行与布料交互的方法。对主流引擎中布料建模方式为物理交互类型的布料模型进行人体的交互试验，完成二者的实时响应过程。

该方法被集成如下的增强现实应用之中：首先对深度采集设备Kinect采集的人体点云数据进行预处理得到建模需要的特征点数据，基于特征点和人体结构学理论来建立由基本几何碰撞体构成的人体模型；再由该碰撞体模型直接完成与预设置的布料进行实时的碰撞检测与响应，最后将渲染得到的布料与真实场景进行融合得到虚拟布料与人体实时交互的效果。

本文对该建模方法进行了理论分析和实验验证，同时对比了其作为增强现实应用与其他虚拟试衣系统在实时性与真实性的实现差异。在建立的增强现实环境下的虚拟试衣系统中进行大量实验，结果表明该方法能够给予用户非常直观的穿衣体验，并能使用户得到逼近真实衣物在自己形体上的展现状态的反馈，而且，直接通过点云建立而成的碰撞体可以加速虚拟试衣算法，实时帧率明显提高。该方法很容易被集成，为实现增强现实下的三维虚拟试衣系统打下基础。

In recent years, virtual fitting technology has been a hot research direction with the development of human-computer interaction. Though virtual fitting technology based on somatosensory are matured enough to show the effect of clothing on users in a variety of positions, the applications based on virtual reality lack of usability, for the reason that users cannot get the distinct effect of the clothes are with theirs real bodies. Augmented reality is of great help of changing the way users feel about virtual fitting while shopping. At now, as a novel research direction, the related technologies of the AR fitting are under investigation. The AR fitting applications have used a solution of combining 2D clothing patch with human body in real scene. Thus, the lack of physical realism problem still remains.

According to this problem, this paper outlined virtual fitting system in augmented reality, and designed a novel method of interaction between human and cloth, which uses collision body model built with the point cloud of a human body to interact with cloth instead of virtual human model. The cloth model used in the whole real-time interaction process was chosen to be the interactive cloth supported by most popular 3D engines.

The method was demonstrated in an AR application. At first, a few feature points is obtained in the pre-process of point cloud of a user. Then a personalized collision model will be created with the feature points based on the structure of human body. By simulating the poses of the user in real time, the model can interact with a pre-defined cloth model in virtual scene, algorithms of collision detection and response has been applied in this step for acceleration. Finally, we merge the interacted cloth into the video frame of real scene as the result.

The feasibility of this method has been proved with both theories and experiments in the paper. And a discussion has been made of the differences between our method and other ones used in popular virtual fitting applications. It shows that our method can give a very distinct feeling of users' wearing the cloth, which means that a user can tell whether a cloth is fit his/her body from the feedback image playing in the screen. And the frame rate is raised by the collision model, for the human body reconstruction process can be omitted. For the simplicity of being integrated, this method can be the foundation of virtual fitting application based on augmented reality.