A Viewer Centric Depth Adaptation System and Architecture Design

碩士 === 國立臺灣大學 === 電子工程學研究所 === 100 === 3D technology had been matured in recent years. There are many researchers and companies work on the depth generation field. Stereo matching is a well developed algorithm can generate depth information from left and right views. Another depth generation that r...

Full description

Bibliographic Details
Main Authors: Chien Wu, 吳謙
Other Authors: 陳良基
Format: Others
Language:en_US
Published: 2012
Online Access:http://ndltd.ncl.edu.tw/handle/48587502562922807192
Description
Summary:碩士 === 國立臺灣大學 === 電子工程學研究所 === 100 === 3D technology had been matured in recent years. There are many researchers and companies work on the depth generation field. Stereo matching is a well developed algorithm can generate depth information from left and right views. Another depth generation that receive high interest is the 2D to 3D algorithm. Although 2D to 3D depth generation is a ill-pose problem. Researchers had solve the problem by using monocular cues to acquire the depth information. All these algorithms are aimed to provide high quality 3D contents. Although we have high quality of depth generation, still there are many users complained about the comfortable issues while watching 3D TV. The main problem is the lack of considering human factors. The viewing condition is various among different users. There may be a change in display type, display size, viewing distance, viewing angle and even the intercular distance between the eyes. And if we don''t consider all these parameter , people would feel some artifacts. We reviewed some of the most important artifacts that interference comfortable viewing experience. Such as shear distortion, card-board effect, puppet theater effect, depth and scale distortion. We then proposed a viewer centric depth adaptation system to reduce these distortions. In this thesis, we focus on algorithm and hardware architecture design of disparity adaptation system. At first, we proposed the depth shifting and remapping method. The idea is to minimize the accommodation-convergence conflict by shifting highest saliency object back to the screen plane. And the depth remapping is applied to compensate the nonlinear perceptual effect. Secondly, we proposed depth scaling with the user condition as input. Our algorithm can provide better cue consistency between relative size and perceived depth. Moreover, the shear distortion is reduced by the concept of virtual display plane. Finally an effiecient VLSI architecture of real-time, high-performance of our algorithms had been proposed. The design combined the depth remapping method, depth shifting and viewing distance based adaptation to reduce magnification distortion and shear distortion. When operating at 227MHz, the architecture can render HD1080p at 30fps.