【作者】 李星；

【导师】 何中市；

【作者基本信息】 重庆大学 ， 工程硕士（计算机技术领域）（专业学位）， 2016， 硕士

【摘要】 本文设计与实现了基于单摄像头的三维虚拟试衣系统,该系统要求用户输入用户的正面和侧面身着紧身衣的照片信息、选择的偏好服装样式,然后经过系统的一系列处理,最终输出与用户身形匹配的三维模特穿上对应的三维服装的穿衣效果,并且穿衣模特能够实现旋转效果,以便能够多角度查看试穿效果。该系统的主要目的是使用户在购买服装时可以通过该系统的试穿体验来查看所选购服装是否合体,样式是否好看,以减少用户在服装网购中因为服装不合身等原因造成的退货问题。本文系统进行虚拟试衣的基本原理是通过单摄像头采集用户身着紧身衣的正面、侧面照片,然后经过图像灰度化、边缘检测以及基于Harris角点检测算法提取出人体的特征尺寸,并进一步生成人体三维模型;同时采用包围盒层次法进行二维衣片的碰撞检测,逐步进行缝合,以模拟三维服装试穿到人体模特身上的效果。本文系统是在Android平台进行设计实现,主要技术内容及工作如下:(1)用户人体特征参数的获取。对采集到的用户原始图像依次进行灰度化处理、Canny算子边缘检测、基于Harris角点检测算法来提取人体图像中的关键特征尺寸信息。在边缘检测过程中,对多种边缘检测算子进行实验对比,最终选择了高信噪比、高定位精度的Canny算子作为人体图像的边缘检测算子。(2)三维人体模特与服装模型的建模。采用3D Studio Max三维建模工具进行人体模型的创建,采用Photoshop等绘图工具绘制服装二维衣片模型。人体模型与服装模型都需要预创建多个,以满足人体与服装相互匹配要求。(3)碰撞检测算法进行服装试穿。采用基于包围盒层次法的碰撞检测算法来进行衣片的缝合操作,同时采用“边-边”、“点-三角形”的非精准碰撞算法优化检测效率,最终达到三维服装穿到人体模特身上的效果。更多还原

【Abstract】 The title of our article is the design and implementation of 3D virtual fitting system based on single camera. Our System requires the user’s photos of the body front and the body side as input, besides the user need to wear tight clothes when he is photographed. When the user chooses his favorite clothes, our system will start to work right now. After our system accumulates a large amount of information, it will sho w the most matched 3D model as the results of our virtual fitting, and the 3D model is matched with the corresponding 3D garment. Our models can rotate, so users can view the results of the virtual fitting in multiple angles. The main purpose of our system is to help users to decide whether it’s appropriate to buy the clothes and let them know whether the style of clothes is suit for them or not. Besides, another purpose is to reduce the return rate in the online clothing shopping.The basic principle of our virtual fitting system is as follows: first, the system will collect the front pictures and side pictures of users, who wear tight clothes. Then after a series of processing like gray image, edge detect ing, corner detecting and so on, we’ll detect corner features. The corner detection algorithm is a classical algorithm proposed by Harris, which works well in detecting corner features. Next, we can extract the feature information of the user’s body and further create a matching 3D model. Finally, we will detect collision of two dimensional garment by using bounding box hierarchy algorithm to suture pieces step by step. So we can simulate the results of 3D virtual fitting.In the paper, our system is implemented on the Android platform, and the main key points are as follows:(1)Collecting the feature parameters of human body. First we’ll collect the user’s original pictures. Then after graying the pictures, we will detect the edges of the pictures by using the Canny operator. Finally, we’ll extract the feature information of the human body by using the corner detection algorithm based on Harris. In the process of edge detection, we select the Canny operator as our main edge detection operator after comparing several edge detection operators because of its’ high SNR and high positioning accuracy.(2)Creating 3D human models and garment models. We use the software named 3D Studio Max to create 3D human body models and the Photoshop software to draw the models of the garment. We need to create multiple models of human body and clothing in advance to meet the requirements of human body and clothing matching.(3)Using collision detection algorithm to virtual fit. We use the collision detection algorithm which is based on bounding box hierarchy to suture the two dimensional garment pieces. Besides, we’ll use the non-accurate collision algorithm based on "edge-edge", "point-triangle" to optimize the detection efficiency. Now the work of 3D virtual fitting is completed.更多还原