A Clustering Based Point Pattern Matching Algorithm In 3D

In the field of computer image registration,the research on point pattern matching occupies an extremely important position,which is the focus of many scholars' research.It is widely used in areas such as image recognition,tracking of moving objects and security.This problem is an NP problem.Due to the wide background of the problem,complex geometric changes and numerous external disturbances,the current research work is still in the exploratory stage.Solving this problem is still a challenging and practical task.This paper analyzes a variety of point pattern matching algorithms,especially focusing on the matching algorithms based on shape context.The core idea of the shape context method is to use the distribution of other points around a point as its characteristics and the overall distribution of the three-dimensional point set as the matching element,which leads to the algorithm being greatly affected by the distribution of abnormal points and not robust.In addition,this algorithm has a poor matching effect on mixed point sets.Therefore,this paper proposes a new 3D point pattern matching method—Gravitational Shape Context based on clustering(GSC).The core idea of the algorithm is to first use the clustering algorithm to split the mixed point set without prior knowledge,and then match the point set to be matched with each cluster of the mixed point set to obtain the best mapping.The innovations of this paper include four aspects:(1)A hybrid algorithm based on the 3sigma principle and Tukey boxplots,which filters the point set and removes outliers.(2)The clustering algorithm is used to split the mixed point set,and the set of points to be matched is matched with each cluster of the mixed point set.(3)Inspired by the universal gravitation formula,the feature extraction method was redefined,and the original feature extraction method for segmenting concentric spheres was simplified.(4)When extracting the features of a point in three-dimensional space,this paper proposes a new calculation method to calculate the spatially invariant points.The weights can be assigned according to the cluster centers of different clusters andthe number of points in the current cluster.Finally,the relatively constant points in space are calculated.In the calculation of the matching degree of the algorithm,this paper proposes a method based on clique matching verification.The original point set to be matched is subjected to rigid transformations such as rotation,translation,scaling,etc.to obtain a new point set,and the three point sets are matched in pairs to find their respective mapped point pairs.Then use the clique matching algorithm to construct an undirected graph between multiple point sets and find common points in all point sets.Consider this common point as the point that eventually matches correctly,and calculate the degree of matching.The experimental data in this paper is simulation data and real data of objects in3 D space.It is used to test the gravity shape context algorithm and multi-point pattern cluster matching verification results proposed in this paper,and compare the proposed algorithm with other point pattern matching algorithms.The experimental results show that the clustering-based gravity shape context model algorithm has a clear advantage over other algorithms in the problem of matching mixed point sets.