Font Size: a A A

GPU Acceleration And Special Hardware Implementation Of3D Television Terminal

Posted on:2014-01-01Degree:MasterType:Thesis
Country:ChinaCandidate:P F JinFull Text:PDF
GTID:2248330395976067Subject:Electronic and communication engineering
Abstract/Summary:
With the development of the computer science, more and more technologies need to use scene3D information to truly represent the objective world. In the past two years,3DTV developed vigorously. More and more scientific research institutes and manufacturers are into3DTV technology research. Naked eyes3DTV with "2D+Depth" coding scheme, renders multi-view images for display according to the depth information. Wearing glasses3DTV system using3D glasses makes the audience get3D perception.In the naked eyes3DTV system, depth information acquisition and rendering the multi-view consumes large amount of computation time. In wearing glasses3DTV system,3D content production also needs to consume large amount of computation time. But real-time is one of the important factors in3DTV system. In order to meet the needs of the consumers, using hardware resources such as GPU, FPGA to accelerate the algorithm and implement the real-time property is a good choice. This paper focuses on the research of GPU acceleration and special hardware implementation of3D television terminal. The main achievements include:(1) A stereo matching algorithm based on two stereo videos is proposed, which is used to obtain dense depth map. In order to look for suitable stereo matching algorithm for natural3DTV system, the use of GPU architecture implements real-time depth information acquisition.(2) A multi-view rendering technology based on image depth is proposed. By the depth information of the reference point image, this technology calculates the corresponding position of each pixel point in a virtual view image. In the shift-sensor camera model, we can obtain the coordinate mapping relationship suitable for the hardware implementation, and implement real-time multi-view rendering using FPGA hardware resource.(3) An algorithm based on2D input video to obtain another way virtual video is proposed. This algorithm can convert traditional2D video into3D video. On the premise of acceptable stereoscopic effect, the use of GPU accelerates the computation speed in order to implement real-time display.
Keywords/Search Tags:stereo matching, multi-view rendering, 2D convert to3D, GPU, FPGA
Related items