3D Reconstruction Of Metal Parts With Duplicate Structures

With the development of machine vision technology,image 3D reconstruction technology has become one of the effective methods for industrial part inspection due to its low cost and easy implementation.However,repetitive structures and similar textures can be found everywhere in practical applications,and such similar and blurred structures can easily lead to the reconstruction of wrong models.Therefore,for parts containing repetitive structures and repetitive scenes,this paper explores the methods that can identify such features and correct the errors caused by them,and furthermore,for the inherent highly reflective and low-reflective characteristics of smooth metal part surfaces,the repair of the holes produced by them is further investigated.The main research of this paper is as follows:Spatial location-based feature point matching method.Due to the influence of similarity features,a large number of incorrect correspondences can easily appear in the2 D matching process,which can affect the subsequent 3D mapping and are difficult to be detected.Therefore,this paper proposes a method to reject the false matches based on the correlation of spatial location of matching points.The method uses cluster relations to filter the 2D matching points that do not belong to the same region in space,and obtains the correct set of matching points,which lays the foundation for the next step of incremental 3D reconstruction.Post-processing algorithm of repetitive structural 3D reconstruction based on conflict metric.In the process of incremental reconstruction,fuzzy features can cause the matching trajectory to fail,and wrong pairwise geometric relationships can make the spatial point position calculation wrong,all these factors will affect the final result.In this paper,we study the post-processing algorithm based on conflict metric to correct the model structure after incremental reconstruction.The camera pairs are first segmented using minimum spanning tree,initially sieved according to the ratio threshold of unique points to duplicate structure points in the camera pairs,and then all false connected phase groups are split using the conflict metric method of super-pixel segmentation.This method can effectively correct the erroneous structures generated by duplicate features.Hole repair methods based on boundary boosting and implicit surfaces.After obtaining the correct point cloud model of a repetitive structural part by the above method,it is obvious that the model has small holes,which is an unavoidable problem for part reconstruction.In this paper,we propose a hole repair method combining boundary boost and implicit surface.After identifying the boundary line inside the hole,the method uses the boundary feature points and their neighborhood points to calculate the filling point direction,and then fills the inside of the hole according to the set moving distance to obtain the initial new point inside,and then constructs an implicit surface based on the radial basis function through the interpolated constraint points and additional constraint points obtained,and transforms the initial new point inside the hole The initial new points inside the hole are iteratively adjusted to this implicit surface to obtain a smooth transition filling model.