Research On Key Technologies Of Point Cloud Data Processing In Industrial Automatic 3D Inspection

With the increasing development of 3D measurement technology and computer technology,it has become the trend of industrial inspection,that use automated device equipped with 3D scanner to reconstruct the 3D model of workpiece,and then analyzed with the CAD model to complete the industrial automatic 3D inspection.However,the data acquired by 3D scanner are generally point cloud that represent the 3D coordinates of workpiece surface,and due to the influence of on-site inspection environment and the like,the point cloud data needs to be processed,and generally includes preprocessing,registration of point cloud and CAD model,surface reconstruction and etc.The point cloud preprocessing step is mainly to eliminate redundant data to improve processing efficiency,reduce distorted data or irrelevant data acquired due to bad factors,and add auxiliary information for point cloud data.Although existing algorithm can obtain good processing result,it usually needs artificially given parameters or takes a long time.It is difficult to achieve the purpose and efficiency of automated inspection.In order to align the measurement point cloud with CAD model for subsequent analysis,the ICP algorithm is usually used.But since the unrelated data obtained during the measurement is difficult to totally remove by preprocessing step,the result of ICP algorithm which is essentially a leastsquare method is not stable.The surface reconstruction process is to transform the point cloud data into a solid model to better express the surface of workpiece,but existing algorithms usually cannot meet the efficiency requirements of industrial inspection.Therefore,although point cloud data processing technology has been widely used in various fields,further research is needed for industrial automatic 3D inspection.Based on previous researches,this paper has conducted in-depth research on key technologies of point cloud data processing,completed the processing steps and ensured the accuracy and efficiency.Here are the main work this paper include:(1)Improved the grid sampling algorithm by introducing the point cloud resolution to realize the simplification process,adaptively specific the grid size without manual.(2)Using the nature of measurement point cloud,an outlier removal algorithm based on statistical variance distance is proposed,which can quikly and effectively remove outliers.(3)Aiming at the unremovable irrelevant data in the preprocessing step,an improved ICP algorithm based on self-adaptive threshold is proposed,which can effectively eliminate the interference of irrelevant data and complete the accurate registration process between measurement point cloud and CAD model.(4)A surface reconstruction algorithm based on restricted Voronoi cells is implemented.This method not only guarantees that the reconstruction result can correctly represent the surface topography of workpiece,but also can achieved in local parallelism to improve the efficiency.Through the verification of actual enginieering projects,the various algorithms proposed and implemented in this paper can quickly and effectively complete the point cloud data processing in industrial automatic 3D inspection.