图像运动估计算法研究与对称多处理系统的研制
文献类型：学位论文
| 作者 | 李相国 |
| 学位类别 | 博士 |
| 答辩日期 | 2009-05-27 |
| 授予单位 | 中国科学院声学研究所 |
| 授予地点 | 声学研究所 |
| 关键词 | 运动估计 并行处理 对称多处理 嵌入式系统 |
| 其他题名 | Research on Image Motion Estimation Algorithms and Development of Symmetric Multiprocessing System |
| 学位专业 | 信号与信息处理 |
| 中文摘要 | 在图像/视频处理应用中，运动估计是一个非常基础的预处理步骤，但它的计算量非常大。因此，对运动估计算法进行并行处理具有重要意义。本文主要研究了基于区域的运动估计算法，并设计开发了一个基于PowerPC的对称多处理系统。本文主要工作及创新有如下几个方面。 首先介绍了基于区域运动估计的三类算法，即块匹配法、梯度法和相位相关法，详细分析了它们的优缺点。对属于梯度法的Keren算法进行了强度扩展，使之能够适应存在强度差异的图像配准；在比较三类算法优缺点基础上，提出了一个结合相位相关法与Keren算法的两步实现算法，它结合了两者的优点，能够在大角度大平移情形下获得高精度的亚像素估计，并通过仿真实验对其有效性进行了验证。 目前的并行处理方案主要包括两类，即基于硬件的并行处理和基于软件的并行处理。前者包括ASIC和FPGA，后者包括基于DSP和通用并行计算平台。由于微处理器性能的极大提升以及多核处理器的发展，故选择基于软件的并行处理方案，选用PowerPC处理器设计一个对称多处理器系统。本文设计了系统的总体框架，详细介绍了各接口模块的原理设计，并介绍了系统的PCB设计及硬件调试。结果显示，系统硬件能够正常工作。 对称多处理系统必须通过操作系统的支持才能正常工作，软件开发包括启动代码、操作系统和应用软件三个层次。本文选择对U-Boot和Linux进行移植构建系统软件。详细介绍了U-Boot和Linux的移植过程和注意事项，并介绍了基于该系统的应用软件开发。 最后，对本文进行了总结，分析了本文的不足和需要改进的地方。 |
| 英文摘要 | Motion estimation is a fundamental preprocessing step for many image or video applications. It consumes high computational cost, consequently parallel processing plays an important role for it. This dissertation mainly discusses algorithms for area-based method, and a symmetric multiprocessing (SMP) system based on PowerPC is developed. The contents and contributions of the dissertation are described in the following. Firstly, three classes of algorithms for area-based method are introduced, i.e. block-matching algorithms, gradient algorithms and phase-correlation algorithms. Their advantages and disadvantages are analyzed in detail. The dissertation extends Keren algorithm which belongs to gradient algorithm to intensity difference. It means that the extended one can deal with images with intensity difference. Based on the analysis of characteristics of these methods, a two-stage algorithm which combines phase-correlation and Keren algorithm is proposed, which inherits advantages of the two methods. This algorithm can reach high precision sub-pixel estimation under large translation and rotational angle situation. Simulation results confirm its availability. Currently, parallel processing schemes are classified into two categories: hardware-based and software-based. The former includes ASIC and FPGA, while the latter owns DSP-based and general parallel computation platform. Due to the micro-processor’s performance advance and its trends for multi-core, software-based scheme is chosen to develop a SMP system based on PowerPC. The top-level frame is designed, and schematic for each interface module is presented in detail. PCB design and troubleshooting for the system are also introduced. It’s shown that this system can work smoothly. A proper operating system is needed for SMP system to manage its resources and let it work properly. The software development for a SMP system includes a boot loader, a proper operating system and applications. U-Boot and Linux are chosen to construct the system envirment. The porting processes and notices for U-Boot and Linux are introduced. Application software development for the system is also included. Finally, a summary is presented for the whole dissertation, limitation and further improvements for the work are pointed out. |
| 语种 | 中文 |
| 公开日期 | 2011-05-07 |
| 页码 | 113 |
| 源URL | [http://159.226.59.140/handle/311008/166]  |
| 专题 | 声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 李相国. 图像运动估计算法研究与对称多处理系统的研制[D]. 声学研究所. 中国科学院声学研究所. 2009. |
入库方式： OAI收割
来源：声学研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。