基于Kinect的虚拟视点生成技术

Virtual view synthesis technology based on Kinect

传统的多视点生成方法是基于多相机阵列系统的关键技术。现提出了基于Kinect的多视点成像计算方法。首先对Kinect的深度图使用三边滤波器进行平滑,根据修复好的深度图配合彩色图,利用DIBR技术生成多个存在空缺信息的彩色视点;最后结合彩色图的纹理结构信息和深度图的背景信息对有丢失信息的彩色图进行修复。实验结果表明,文中提出的深度修复方法能够有效地修补Kinect的深度图,生成的虚拟视点图在3DTV上效果明显,立体视觉效果显著。

Conventional multi-view image is a key technology based on multi-camera, which are always expensive to setup and cumbersome to use. The paper proposed a light weight multi-view system via a computational approach with Microsoft Kinect. Missing areas in depth map are pre-processed by a trilateral filter which is guided by edge information above all. Inpainted depth map combined with corresponding color map is used to synthesized virtual views. The lost background information in raw views are inpainted by text structured information and priority determined by depth map. The experimental result show that the method restores high quality depth maps and strong 3D visual experiences are observed when synthesized views are shown in 3DTV.