非常规摄影条件下立体像对的同名点匹配方法研究

发布时间：2018-09-09 20:41

【摘要】：图像匹配技术是将不同视角、不同传感器、不同时间下获取的具有一定重叠部分的两幅影像进行匹配，它是图像处理的一个基本问题。图像匹配方法大致可以分为两类：一、基于灰度的图像匹配方法，二、基于特征的图像匹配方法。本文总结介绍了前人在图像匹配领域的研究成果及现状，详细介绍了图像匹配的流程，并且提出了极大似然估计（MLESAC）算法用于非常规摄影条件下立体像对的同名点匹配当中。在研究什么是非常规摄影条件下，介绍了沿主光轴方向的摄影，以及大转角的摄影，，此类情况下均是非常规摄影。本文详细介绍了，沿主光轴方向的摄影，画出了其立体像对的成像模型，也画了常规摄影中立体像对的成像模型，从图中可以看出沿主光轴方向的摄影的视差、核线排列等问题与传统的常规摄影测量有很大的区别。在研究基于特征的图像匹配方法时，本文提出了基于SIFT（尺度不变特征转换）算法并加入极大似然估计(MLESAC)算法用于非常规摄影条件下立体像对的同名点的匹配,用SIFT提取特征点，再用基于似然函数值的极大似然估计算法(MLESAC)去除误匹配的特征点对，实现图像的精确匹配。并用随机采样一致性算法(RANSAC)对SIFT提取的特征点对进行处理对比分析了两种方法处理后的结果。实验验证，MLESAC算法应用于非常规摄影条件下立体像对的同名点匹配是高效稳定的图像匹配。最后对本文所做的工作进行了总结，并对本文在图像匹配领域有哪些需要深入研究的地方进行了展望。
[Abstract]:Image matching is a basic problem in image processing, which is to match two images with overlapped parts obtained from different angles of view, different sensors and different time. Image matching methods can be divided into two categories: one is gray-based image matching method, the other is feature-based image matching method. This paper summarizes and introduces the research achievements and current situation of previous researches in image matching, introduces the flow of image matching in detail, and proposes a maximum likelihood estimation (MLESAC) algorithm for matching stereo pairs with the same name under unconventional photography conditions. Under the condition of studying what is unconventional photography, this paper introduces the photography along the main optical axis, as well as the photography with large rotation angle, in which case the photography is unconventional. This paper introduces in detail the imaging model of the stereo image pair along the main optical axis and the imaging model of the stereo image pair in the conventional photography. The parallax of the photography along the main optical axis can be seen from the picture. Nuclear alignment and other problems are quite different from conventional photogrammetry. When studying the feature-based image matching method, this paper proposes an algorithm based on SIFT (Scale-Invariant feature Transformation) and adds the maximum likelihood estimation (MLESAC) algorithm to match the points of the same name of stereo image pair under the condition of unconventional photography. The feature points are extracted by SIFT. Then the maximum likelihood estimation algorithm (MLESAC) based on the likelihood function is used to remove the mismatched feature pairs to realize the accurate image matching. A random sampling consistency algorithm (RANSAC) is used to deal with the feature pairs extracted by SIFT. The results of the two methods are compared and analyzed. Experimental results show that the MLESAC algorithm is an efficient and stable method for stereo image pair matching with the same name. Finally, the paper summarizes the work done in this paper, and looks forward to what needs to be further studied in the field of image matching in this paper.
【学位授予单位】：中国地质大学（北京）
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.41;P23

【参考文献】