摄像机标定及立体匹配技术研究

发布时间：2018-04-13 15:22

本文选题：标定 + 稠密匹配　；参考：《南京理工大学》2017年硕士论文

【摘要】：立体视觉的实现可分为摄像机标定、特征点提取、立体匹配、三维信息的恢复及后处理五个步骤。本文围绕计算机立体视觉系统,针对摄像机标定和立体匹配技术展开深入研究,为实现从二维图像中提取三维信息打下基础。在摄像机标定部分,本文采用了综合传统标定和自标定优点的张正友平面模板标定法获得了双目摄像头的内外参数。同时对双目图像进行了畸变校正和极线校正,改善了图像质量并使双目摄像头拍摄的左右成像平面对极线行对齐,为后续图像立体匹配提供了方便。在特征点初匹配中,本文对几种常用的特征提取算子性能进行对比并选择了抗噪性和稳定性较好的SIFT算法进行特征提取和匹配,同时采用RANSAC算法去除误匹配点,从而达到了精匹配效果。在特征点稠密匹配中,本文首先研究了基于区域增长的稠密匹配方法,详细阐述了具体实现过程。同时本文提出了一种基于单应性的稠密匹配方法,该方法通过不断地假设当前的配准结果中相互毗邻的三对特征点形成的三角面片满足单应性关系,并利用互相关函数对单应性假设进行校验。将符合校验原则的三角面片记为配准面片,将不满足校验原则的三角面片进行细分后重新判断,既而从图像上检测更多的稠密匹配点,且匹配准确度较高。最后本文对两种稠密匹配方法的实验结果进行对比,验证了该方法的有效性。
[Abstract]:The realization of stereo vision can be divided into five steps: camera calibration, feature point extraction, stereo matching, 3D information recovery and postprocessing.In this paper, the camera calibration and stereo matching technology are deeply studied around the computer stereo vision system, which lays the foundation for extracting 3D information from two-dimensional images.In the camera calibration part, the internal and external parameters of the binocular camera are obtained by using the Zhang Zhengyou plane template calibration method, which combines the advantages of traditional calibration and self-calibration.At the same time, the distortion correction and pole line correction of binocular image are carried out to improve the image quality and align the left and right imaging plane of binocular camera, which provides convenience for stereo matching of subsequent images.In the initial matching of feature points, this paper compares the performance of several commonly used feature extraction operators, and selects the SIFT algorithm with good noise resistance and stability for feature extraction and matching. At the same time, RANSAC algorithm is used to remove the mismatch points.Thus, the precision matching effect is achieved.In the dense matching of feature points, this paper first studies the dense matching method based on regional growth, and describes the implementation process in detail.At the same time, a dense matching method based on homotropy is proposed. This method continuously assumes that the triangular patches formed by the three pairs of feature points adjacent to each other in the current registration results satisfy the monotropic relationship.The hypothesis of homotropy is verified by cross-correlation function.After subdividing the triangulated face which does not meet the checkout principle, more dense matching points are detected from the image, and the matching accuracy is high.Finally, the experimental results of two dense matching methods are compared to verify the effectiveness of the proposed method.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】