Research On 3D Reconstruction Of Casting Based On Monocular Multi-viewpoint Image

In order to realize automatic detection and recognition of casting burrs based on machine vision,a 3D reconstruction method of castings based on monocular multi-viewpoint images is proposed.This method has a very wide range of applications in computer vision,industrial automation,machine vision testing and other fields.A smartphone is used to rotate around a stationary casting which with burr for multi-view capture to get a sequence of images.The image sequence of the group as a 2D image source for 3D reconstruction method.The specific research contents include: Firstly,extracted and matched the feature points from every two adjacent images.Aiming at the problem that there are many mismatched points in the rough matching of SIFT(Scale Invariant Feature Transform),a two-step matching method is proposed in this paper.Combining the distance measure method and the similarity function method,and then we add vector space cosine similarity constraint conditions on the basis of Euclidean distance to exclud false matching point pairs.We can eliminate 76% of the mismatched point pairs,compress the redundant data and improve the efficiency of the subsequent 3D reconstruction.Secondly,the feature point is reconstructed by sparse point cloud using SFM(Structure from Motion)algorithm.The algorithm does not need to calibrate the camera in advance.Only need to analyze the geometric relationship between the multiple images and the corresponding relationship between the feature points to calculate the camera's position information and 3D sparse point cloud model of the casting.Then,the dense point cloud is densely reconstructed by multi-view matching of sparse point clouds.Finally,we reconstruct the dense point cloud by surface texturereconstruction.In this process,Poisson surface reconstruction algorithm is used to convert the surface reconstruction of point cloud data to solve the Poisson equation to obtain the best watertight closed feature fitting the surface and finally obtaining the 3D reconstruction model of the target object with good surface features and detail features.Forty-four image sequences were obtained by monoscopic visualization of the castings which with burr and were reconstructed in three dimensions.Experimental results show that the method proposed in this paper can improve the efficiency of3 D reconstruction,reconstruct the details of 3D surface and reconstruct the surface smoothness.Monocular vision device is simple,low cost,small space required,without pre-calibration of the relative position of the camera and the target,suitable for three-dimensional surface reconstruction of parts under mechanical processing environment.