Subaperture Stitching Algorithm Research On Structured Surfaces Measurement

With the development of current technology,the rise of micro-nano devices is becoming more and more possible,which are widely used in automotive,aerospace and other fields.There are many kinds of geometric characteristic structures on the surface of microstructural elements,such as micro-steps,micro-grooves,etc.Their geometric characteristic structures have an important impact on the microstructural elements,so it is necessary to measure and evaluate the microstructural elements.However,due to the limits of the resolution of the image sensor and optical magnification,the measurement range of all surface topography measuring instruments is limited.Most of the large-scale surface topography of microstructures across the scale must be measured by piece-wise sub-aperture,and then the stitching calculation is used to realize the stitching of the whole data.Therefore,stitching algorithm is a very important factor affecting the accuracy of the surface topography of the measuring object.Traditional three-dimensional matching algorithms usual y use key feature points and their corresponding neighbors information to construct feature descriptors.Because the structure of structured surface is arranged in the form of arrays,the neighbor of feature points on most of surfaces is very similar,so that the general three-dimensional matching algorithms are prone to generate the matching ambiguity and can not accurately solve the matching problem of structured surfaces.In this paper,a three-dimensional stitching algorithm based on feature points clustering is proposed.The simulation and experiment verify that it can be used for accurate stitching of sub-aperture measurement data of structured surfaces.The algorithm consists of two parts: searching for inlier group feature points pairs(correct feature matching pairs in overlapping regions)and calculating three-dimensional rigid transformation.In the search of(twodimensional)inlier group feature point pairs,firstly,SIFT algorithm is used to detect feature points in surface data for arbitrary adjacent sub-aperture data.Our SIFT feature analys is experiments on surface topography data show that this method has a good immune effect on data changes caused by the rotation or position offset of samples in measurement.Then,mean shift clustering algorithm is used to classify the extracted feature points,and the repetitive feature points and salient feature points are obtained by the proposed classifying principle.In order to avoid the fuzziness of searching corresponding point pairs,the least squares homography transformation calculation and outlier feature points pairs elimination are carried out by randomly selecting different salient feature points pairs,and the verification of homography transformation is carried out by using repetitive feature points pairs to make the algorithm converge to an optimal solution.The solution guarantees the minimal sum of squares of distances between salient and repetitive feature pairs in sufficient number,and gives the nonambiguous feature pairs.Finally,the Z coordinates of all feature points pairs are introduced to calculate the three-dimensional rigid transformation matrix between all inlier group point pairs,and it is used for the stitching of sub-aperture data.Through a series of simulation and comparison experiments,it is proved that the proposed sub-aperture stitching algorithm has better robustness and efficiency for the stitching of structured surface topography data.