【作者】 刘红；

【导师】 龚卫国；

【作者基本信息】 重庆大学 ， 仪器科学与技术， 2012， 硕士

【摘要】 随着计算机技术和信息科学的飞速发展，图像匹配技术的应用越来越广泛，而对其性能的要求也不断提升。目前图像匹配技术的难题主要为以下两个方面：计算速度达不到实时性要求；在遮挡、大尺度缩放、光照变化、旋转、模糊等情况下的鲁棒性不够高。为此，本文展开了图像匹配技术相关的研究工作，改善了算法的速度和鲁棒性。主要研究内容可概括为以下几点：①对图像匹配技术的研究现状进行了调研，为下文的算法研究奠定理论基础。介绍了图像匹配技术的基本流程，简述了图像预处理中图像灰度化的处理方法，阐述了误匹配约束RANSAC算法的相关理论基础，并列出了图像匹配领域常用的实验图像测试库以及实验评价标准。同时还在理论准备阶段研究了摄像机标定方法，并选取了标准方格的标定方法对摄像机进行了标定实验。以上工作都为快速图像匹配技术的研究铺垫了坚实的基础；②研究了一种基于DAISY描述符的图像匹配算法。该算法以FAST角点检测算法和BBF近似搜索算法为基础，采用最新的DAISY描述符进行特征描述，实现了快速鲁棒的图像匹配算法。进行了大量的实验，分别验证了FAST算法、DAISY算法的优秀性能，并将本文设计的新的匹配算法与目前最流行的SIFT算法进行了对比实验，从主观评价和客观评价出发，证明了本算法的高效性和实用性；③针对图像匹配算法在实际应用系统中达不到实时性要求的难题，本文提出了一种并行快速特征点匹配算法。算法采用两步式由粗到精的匹配算法，并利用计算机双核技术创建地图绘制的并行线程，实现快速的并行两步式特征点匹配算法。实验结果表明，该算法成功地应用到实时区域跟踪领域，其运算速度和匹配性能都能满足实时性要求，且在大尺度缩放、遮挡、模糊、旋转等难题下能够鲁棒运行，为匹配技术提供了新的思路。更多还原

【Abstract】 With the rapid development of computer technology and information science,image matching techniques are widely used, and its performance requirements are alsorising. At present, the problems of image matching techniques are mainly in thefollowing two aspects: the calculation speed is less than real-time requirements; therobustness is not high enough under block, large-scale changes, illumination changes,rotation changes, blur changes and so on. So we launched the research on the imagematching technology, to improve the speed and robustness of the algorithm. The maincontents can be summarized as the following:①This paper introduces the present state of the research in image matchingtechnology, which is the theoretical basis for the below algorithm. We introduced thebasic process of image matching techniques, outlined the approach of the imagepre-processing of grayscale and the theoretical basis of mismatch constraint ofRANSAC algorithm, and listed the experimental images test library in the field ofimage matching, as well as experimental evaluation criteria. We also studied on cameracalibration method in the theoretical preparation stage, and selected the standard gridmethod to do the camera calibration experiments. The above work builds a solid basefor the research of fast image matching technology.②We realized an image matching algorithm based on DAISY descriptor. Based onthe FAST corner detector and BBF approximate search algorithm, we used the latestDAISY descriptor to achieve a fast and robust image matching algorithm. A lot ofexperiments were done, respectively, to verify the excellent performance of the FASTalgorithm, the DAISY algorithm. And our new matching algorithm is compared with theSIFT algorithm which is currently the most popular experiments, from a subjectiveevaluation and objective evaluation, to prove its high efficiency and practicality.③For the challenge of the real-time requirements in the practical application ofimage matching algorithm, this paper presents a fast parallel matching algorithm basedon feature points. The algorithm uses a two-step coarse to fine matching method, andtakes advantage of the computer dual-core technology to create the parallel thread ofmapping, to achieve parallel two-step matching algorithm based on feature points.Experimental results show that our algorithm can be successfully applied to real-timeenclosures tracking, computing speed and matching performance can meet the real-timerequirements, and under large-scale view changes, block, blur changes and rotation changes, our method can run robustly. Our research provides a new way of thinkingabout the image matching.更多还原