导　　师： 姚素英

学科专业： 080903

授予学位： 硕士

作　　者： ;

机构地区： 天津大学

摘　　要： 图像拼接是图像处理研究领域的一个新的热点问题。随着人们对图像质量的要求的日益提高，单一传感器获取的图像对于某些特殊要求已经较难满足，故而出现了多传感器系统来满足这些特定要求。图像拼接是多传感器系统中十分重要的一种算法，它能将两幅或多幅有交叠的图像合成成一幅广视角的全景图像，广泛应用于社会的各个领域。 针对图像拼接的需要，本文提出一种基于角点与频谱比对的图像拼接算法，并对实现该算法的SoC中的关键技术进行了研究与实现。该SoC能将两幅约15/%重叠率的图像进行高质量的匹配与拼接操作。 本文首先对图像拼接的流程进行了大体介绍，而后对现有的、流程中涉及到的算法做了简要介绍，并分析了各个算法的优缺点。针对这些优缺点，本文提出了一种适于SoC实现的图像拼接处理算法。其后对算法进行了硬件化，提出了图像拼接SoC与算法模块IP的架构。针对算法中一些较难硬件实现的技术难题，本文通过应用FIFO缓存、流水时序控制等方法一一攻克。同时在系统设计时综合考虑了系统实现的硬件开销与系统的处理运行速度，通过模块复用在保证处理速度的前提下将系统的开销降到了最低。本文提出的SoC基于Xilinx LX110TFPGA开发系统实现，应用了系统提供的MicroBlaze核心作为主处理器进行数据的调度。同时通过对开发系统SystemACE IP修改实现了从CF存储卡读写图像数据的功能。 在软件算法与硬件平台设计完成后，本文对算法与硬件系统进行了仿真与验证。通过软件仿真结果可以看出，本论文提出的算法能对不同尺度变换的图像精确的选出三组匹配点进行图像映射矩阵的计算。算法最终完美的将两幅尺度差异约10/%的图像进行了拼接。通过时序仿真与硬件综合结果，本文设计的硬件可以较为快� Image mosaic is the new hotspot in the image processing research. At the presentage, people’s demand for image quality is increasing. The images captured by oneCMOS sensor failto meet some special requirements. As a result, the multi-sensorimaging system comes into our life. Image mosaic is one of the most significantalgorithms in the multi-sensor system. Through it, two or more small pictures withoverlapped area can be stitched together into one wide view panoramic image. Focus on the need of image mosaic, this thesis provides an image mosaicalgorithm based on the corner points and the spectrum comparison. And the keytechnologies in the image SoC are researched and implemented as well. The mosaicSoC can join two pictures with15/%overlapped area into on image. In this thesis, the steps of image mosaic processing are introduced at first. Thenthe thesis illustrates the algorithms in these steps briefly. After comparing theadvantages and defects of these algorithms, a new mosaic algorithm for hardwareimplementation is given. Then the circuit of this algorithm is designed as well as theSoC architecture. To overcome the difficulties in hardware system design, FIFO andpipeline are used to control the timing. Considering the hardware consumption andthe system speed, this system is designed in a compromising way. Taking theadvantage of module reuse, the hardware consumption is in a low level. The SoCprovided in the thesis is based on the Xilinx LX110T development kit and usingMicroBlaze as the central processor to dispatch the image data. The SystemACEmodule is adapted to access the image data in the CF card. Through the simulation from Matlab, the algorithm which is given in this thesiscan select out3pairs of corresponding point in different scale transform. At last, twoimages with10/%difference are joined into one image. Through the hardwaresimulation and synthesis report, the SoC can fuse two images quickly and precisely.

关 键 词： 图像拼接 特征点检测 频谱比对 片上系统

领　　域： [自动化与计算机技术] [自动化与计算机技术]