面向嵌入式应用的二维视频转换为三维视频的关键技术研究

发布时间：2018-02-26 09:44

本文关键词： 2D转 3D深度估算 DIBR 硬件设计　出处：《天津大学》2012年硕士论文　论文类型：学位论文

【摘要】：近年来,立体成像和显示技术得到了高速的发展,市场上已经出现多种立体显示设备,观察者可以通过众多途径体验到立体效果。然而,立体成像所需的立体片源由于拍摄校准难、后期处理复杂以及拍摄成本高等多种因素限制,导致其数量非常匮乏,从而使得平面视频转立体视频技术受到越来越多的青睐。目前,主流平面视频转立体视频技术存在算法适用范围窄、对移动设备支持少的缺点,影响了全自动视频立体化的实用性。针对上述缺点,本文首先设计了一种基于视频内容分类的全自动深度估计算法:通过对视频运动类型的分析,调用“运动中获取深度”或者“运动中获取结构”获取运动矢量,再通过融合运动矢量和基于颜色的图像分割结果优化深度图,最后使用非对称双边滤波进行后处理。相比现有方案,该算法在保持深度图估算效果的同时,所能处理的场景大幅增加,增强了全自动平面视频转立体视频技术的实用性;其次,通过对立体成像原理的分析,提出了一种适合于硬件实现的基于深度图的立体成像技术(DIBR),并通过数学优化、切割流水线、设计专用缓存等方法大幅提高了处理速度、减少了硬件开销。使用Verilog语言在Xilinx Virtex-5 FPGA上进行了布局布线后表明:所设计的方案硬件资源占用率为现有算法1/30,速度为现有算法的1.8倍。
[Abstract]:In recent years, stereo imaging and display technology has been developing rapidly. There have been many stereo display devices on the market. Observers can experience stereoscopic effects in many ways. However, Due to the difficulty of camera calibration, complex post-processing and high shooting cost, the amount of stereo source for stereo imaging is very scarce, which makes planar video to stereo video technology more and more popular. At present, the mainstream planar video conversion stereo video technology has the shortcomings of narrow range of algorithm application and less support for mobile devices, which affects the practicability of fully automatic video three-dimensional video. In this paper, we first design an automatic depth estimation algorithm based on video content classification: by analyzing the video motion type, we call "get depth in motion" or "get structure in motion" to obtain motion vector. Finally, the depth map is optimized by merging motion vector and color based image segmentation results. Finally, asymmetric bilateral filtering is used for post-processing. Compared with the existing scheme, the algorithm not only maintains the effect of depth map estimation, but also preserves the effect of depth map estimation. The scene that can be processed increases greatly, which enhances the practicability of automatic plane video to stereo video technology. Secondly, through the analysis of stereo imaging principle, A kind of stereo imaging technology based on depth map is proposed, which is suitable for hardware implementation. The processing speed is greatly improved by mathematical optimization, cutting pipeline, designing special buffer, etc. The hardware cost is reduced by using Verilog language in Xilinx Virtex-5 FPGA. The result shows that the hardware resource utilization of the proposed scheme is 1 / 30 of the existing algorithm, and the speed is 1.8 times of that of the existing algorithm.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41;TP368.1

【参考文献】