基于Kinect的多视点立体成像系统-信息与通信工程专业论文.docxVIP

ABSTRACT Multi-view image technology has been in an increasing number of application of computer vision. Traditional multi-view image is based on multi-camera array which is large and heavy to operate. As the depth camera’ application is spreading, an method combining an depth camera with a color camera has attracted more and more attention. The paper introduce the stereo imaging and display technology, as well as the development of abroad at the first part. Then we propose a multi-view system based on Microsoft Kinect. The depth map generated by Kinect is filtered by an edge-guide propagate trilateral filter, and inpainted depth map combined with corresponding color map are used to generate virtual views via depth image based rending technology. The raw virtual views exist large missing areas in background so that we proposed a depth-guide priority method for searching block for inpainting first and the use the matrix completion and matrix recovery model to solve the final problem. The main contributions are listed below: The paper proposed a multi-view image system based Microsoft Kinect RGB-D camara which is different from traditional stereo system. Before the depth map is used to generate virtual views, it is preprocessed by an edge-guide propagate trilateral filter which takes account of color、depth、spatial distance information. The background and foreground is perfectly segmented and missing fields are inpainted. An block priority calculation method for virtual views inpainting is proposed. The method take advantage of depth information at different plane and texture、 structure of color map. As the priority block is searched out along the border of missing area, we consider the problem as matrix completion and matrix recovery so that an Enhanced Lagrangian operator is made use to inpaint virtual color map. The final inpainted results are presented in two kinds of 3D displays, one is two-view stereoscopic display and another is auto-stereoscopic display. Stron