Research On 3D Reconstruction And Point Cloud Registration Technology Of Complex Parts

Complex castings and forgings such as aircraft landing gear,steam turbine blades,and high-speed train wheels are key components in the fields of aviation,ships,and transportation.Their shapes are complex and the forming accuracy is difficult to control.Complete 3D measurement and accuracy analysis of their overall shapes can provide fundmental measurement data for the optimization of the forming process,and is crucial for improving the forming accuracy and improving the ratio of qualified products.Among them,the 3D reconstruction accuracy,and the 3D registration accuracy between the reconstruction result and the design model directly determine the reliability and accuracy of the final quality inspection results.To this end,this dissertation studies the 3D reconstruction method of complex parts,studies the coarse registration and fine registration method between the reconstruction result and the design model.The main research work and innovations are as follows:Research on high-precision 3D reconstruction methods for complex parts.Existing industrial parts 3D reconstruction methods rely on external assistances to locate the measurement viewpoints,which limits the flexibility and application scenarios.Aiming at the problem,this dissertation proposes a high-accuracy 3D reconstruction method based on sequential registration of measurement data and global optimization of measurement viewpoints.The accuracy of sequential inter-frame registration is improved with a novel fast depth images registration method which fuses the geometric consistency constraint and curvature consistency constraint together.The relationship between point cloud registration error and spatial pose error is deduced,the multi-view point cloud registration problem is modeled in the pose graph optimization framework to achieve fast global optimization of the measurement viewpoint poses.The experimental results show that the proposed method effectively eliminates the point cloud data inconsistency caused by the accumulated error in the sequential registration process,and achieves high-accuracy 3D reconstruction of complex parts.Research on the point cloud coarse registration method for two partially overlapped point clouds.A method of constructing a local reference coordinate system with spatial rotation and translation invariance is proposed.The 3D local features are extracted based on the constructed local reference coordinate system.On this basis,the 3D local feature matching is performed on the pairwise point clouds.Aiming at the mismatches in the feature matching results,a geometric-consistency based sample consensus algorithm is proposed.By constraining the Euclidean distance between 3D points in different point clouds,the coarse registration of point clouds is robustly estimated.Aiming at the difficulty to verify the correctness of the coarse registration results,a global-local rigid transformation cross-checking method is proposed,which checks the global rigid transformation from the overall shape registration and the local rigid transformation from the local shape registration,to achieve fast and accurate tests of the correctness of the point cloud coarse registration results.The experimental results show that,compared with the traditional method,our method has high registration correctness ratio,and is robust to different influence factors such as different overlapping ratios,point cloud noise,density variation and outliers.Research on parameterless point cloud fine registration method.Aiming at the problem that the current point cloud fine registration methods need to manually set the algorithm parameters to adapt to different registration conditions,an adaptive distance threshold based iterative closest point(ICP)algorithm is proposed.The adaptive distance threshold is obtained by statistical analysing the corresponding point distances in the ICP coarse registration iterative process,which solves the problem that traditional methods need to adjust to appropriate threshold parameter and their registration results are easy to fall into local optimum.Based on the ICP coarse registration results,the overlapping ratio of pairwise point clouds is estimated and used as the trimming ratio for the mismatching removal in ICP fine registration process,which ensures the final accuracy of point cloud fine registration.The experimental results show that the proposed method does not need to manually adjust the algorithm parameters,and is robust to different types of industrial parts point cloud data which have different overlapping ratios and initial relative positions.Also,our method has the advantages of high registration efficiency and high registration accuracy.The proposed algorithms are applied to the online quality inspection of hot forged automobile front axle parts and high speed train wheel parts.According to the different quality inspection requirements of different industrial parts,the automatic coarse and fine registration are performed to further analysis the overall profile and key dimensions on the front axle hot forging and high-speed train wheel production lines.The proposed algorithm achieves good results in industrial application scenarios,and can meet the requirements of registration efficiency and accuracy for online quality inspection of industrial parts.