基于同步相机阵列的自动人体三维重建

发布时间：2018-08-06 08:40

【摘要】：运动捕捉(Motion Capture)是计算机视觉领域的基础研究课题之一,旨在基于多个不同视角的同步相机阵列恢复人体运动过程中的三维模型和细节特征,在虚拟现实(Virtual Reality)、数字动画、影视制作等领域有着广泛应用。当前主流的基于同步相机阵列的运动恢复虽然在三维模型重建方面取得了一些较好的成果,但主要存在三方面的问题:普通设备无法进行多个视角连续高速捕获的数据获取需求;不能够实现自动分割,需要手动抠取大量的Mask作为输入;往往受制于内存大小限制而无法获得高精度的重建效果。本文基于浙江大学人体反射场高性能捕获系统,通过其同步高速采集的视频序列,进行运动人体三维模型重建。首先,我们设计并搭建了一套完备的基于同步摄像机阵列的自动人体三维重建系统,涵盖了数据采集和标定、自动分割、构建VisualHull可见外壳、深度恢复、点云采样和三维模型重建等模块。其次,提出了一个结合颜色、纹理、对比度和边缘等多线索的自动前景分割方法:先在灰度空间利用背景差分进行前景初始化,其次用Graph Cuts得到初步的分割结果,最后应用Background Cut降低背景区域的对比度,将三种方法的优点很好地结合起来。实验结果表明,我们设计搭建的软件系统合理有效,不仅能够恢复出高精度的深度结果和获得密度可控的点云采样,而且能够稳定地重建出细节突出、表面平滑的人体运动三维模型。所提出的多线索分割算法,相比其他分割方法,能够有效降低光照和前背景颜色相似对分割的影响,实现了高质量的前景自动分割,为自动的人体三维模型重建奠定了基础。
[Abstract]:Motion capture (Motion Capture) is one of the basic research topics in the field of computer vision. It is designed to restore three-dimensional model and detail features in the process of human motion based on the synchronous camera array of multiple different perspectives. It is widely used in the fields of virtual reality (Virtual Reality), digital animation, shadow production and so on. Although the motion recovery of the step camera array has achieved some good results in the reconstruction of 3D model, there are three main problems: the ordinary device can not carry out the data acquisition demand from multiple views and continuous high-speed capture; it is not able to realize automatic segmentation, so it is necessary to manually pick a large number of Mask as input; often it is subject to memory. In this paper, based on the high performance capture system of the human reflection field of Zhejiang University, this paper reconstructs the 3D model of the moving human body through the video sequence collected synchronously and high-speed. Firstly, we designed and built a complete set of automatic 3D reconstruction system based on the same step camera array. It includes data acquisition and calibration, automatic segmentation, construction of VisualHull visible shell, depth recovery, point cloud sampling and 3D model reconstruction. Secondly, a multi clue automatic foreground segmentation method, which combines color, texture, contrast and edge, is proposed. First, foreground is initialized with background difference in gray space, followed by Gr APH Cuts obtains the initial segmentation results, and finally uses Background Cut to reduce the contrast of the background area, and combines the advantages of the three methods well. The experimental results show that the software system we designed is reasonable and effective, not only can restore the high precision of the depth of the fruit and obtain the density controlled point cloud sampling, but also can be used. The proposed multi cues segmentation algorithm, compared with other segmentation methods, can effectively reduce the influence of light and background color similarity on segmentation, and realize high quality automatic segmentation of foreground, which lays the foundation for the reconstruction of self moving 3D model of human body.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】