基于Kinect的实时深度提取与多视绘制算法

提出了一种基于Kinect的实时深度提取算法和单纹理+深度的多视绘制方法。在采集端,使用Kinect提取场景纹理和深度,并针对Kinect输出深度图的空洞提出一种快速修复算法。在显示端,针对单纹理+深度的基于深度图像的绘制(DIBR,depth image based rendering)绘制产生的大空洞,采用一种基于背景估计和前景分割的绘制方法。实验结果表明,本文方法可实时提取质量良好的深度图,并有效修复了DIBR绘制过程中产生的大空洞,得到质量较好的多路虚拟视点图像。以所提出的深度获取和绘制算法为核心,实现了一种基于深度的立体视频系统,最终的虚拟视点交织立体显示的立体效果良好,进一步验证了本文算法的有效性。本文系统可用于实景的多视点立体视频录制与播放。

Real-time depth extraction and multi-view rendering algorithm based on Kinect

This paper proposes a real-time depth extraction algorithm based on Kinect and a multi-view rendering algorithm based on single color plus depth.At the acquisition side,we acquire color information and depth information by Kinect,and present a fast inpainting algorithm to inpaint holes in depth map output by Kinect.At the display side,to solve the big hole filling problem of depth image based rendering(DIBR) based on single color plus depth format,this paper presents a DIBR algorithm based on background estimation and foreground segmentation.Experiments show that our method can get good depth image in real-time,and effectively inpaint the big holes due to DIBR,so multiple virtual view images with good quality can be got.At last we implement a multi-view stereo video system based on the depth extraction and rendering algorithm.A good stereo perception can be achieved by interlacedly displaying the virtual views on auto-stereoscopic display,which further verifies the effectiveness of our method.This system can be used to record and display multi-view stereo videos for live scenes.