基于结构相似度的飞机油箱内窥图像配准方法

Aircraft fuel tank image registration algorithm based on structural similarity

针对飞机油箱隔舱式的连通结构中遍布管、阀、桁条等部件,造成相邻图像匹配困难的问题,研究了一种基于结构相似度的图像配准方法,以解决飞机油箱内窥形貌全景图像拼接问题。该方法首先进行尺度不变特征变换(SIFT)特征点的检测和初匹配,并根据匹配特征点对间的几何特征关系进行初步筛选,然后基于结构相似度的计算找出关键匹配点对,求取最优单应矩阵的模型参数,从而实现飞机油箱内图像拼接。实验表明,该方法能够剔除明显的误匹配点对,增加飞机油箱内窥图像中不同平面结构上的内点,解决了特征点对误匹配、漏匹配的问题,配准效率和评价指标都有了明显的改善。

For the adjacent image matching is difficult due to the connected structure of the aircraft tank full of parts such as pipes, valves and beams, an image registration method based on structural similarity was studied in order to solve the key problem of panoramic image mosaic of aircraft fuel tank endoscope image. The detection and initial matching of SIFT（ ScaleInvariant Feature Transform） feature points were carried out, and the geometric feature relations between feature points were preliminarily screened. Then, the similarity calculation was used to find the key matching points. The model parameters of the optimal homography matrix should be used to realize the image stitching in the aircraft tank. The experimental results show that the method can eliminate the obvious mismatching point pairs and increase the interior points of the different planar structures in the endoscopic image of the aircraft tank, avoid mismatching and leakage matching of feature points, and obviously improves registration efficiency and evaluation indicators.