Research On Pose Estimation Of Workpiece Based On Structured Light

At present,the domestic prefabricated building is mainly steel structure,its components are not standard workpiece and the production line needs high flexibility,which makes the automatic welding technology of steel structure workpiece of great significance,not only can solve the problem of labor shortage,but also can overcome the influence of non-standard parts,greatly promote the development of prefabricated building.However,due to the large size of the workpiece,most of them are assembled and positioned manually,and the assembly deviation leads to the failure to detect welds in the weld positioning process,so it cannot work normally.Therefore,it is necessary to estimate the position and pose of the workpiece to obtain the accurate position and pose,and provide the initial weld coordinate for the weld positioning to ensure the smooth progress of automatic welding.In view of pose estimation of large-size,weakly textured and simple workpiece,Wu Kong structured light camera was selected by comparing two depth sensors,and multiple point clouds were collected based on mobile conveyor belt for stitching to obtain the complete point cloud of the workpiece,and the pose estimation was completed by registration with the model point cloud.Firstly,the accuracy of structured light camera is tested and the internal and external parameters are determined in this paper.The distance between two parallel planes in the collection point cloud was calculated to test the camera accuracy.By taking the camera coordinates of pixel corners to calculate the distance between corners to test the accuracy of internal parameters and determine the internal parameters.External parameters are calculated by aligning the same set of corners in the camera coordinate system and the world coordinate system to complete the calibration of external parameters,and the accuracy of external parameters is tested by converting other feature corners.Secondly,a point cloud stitching algorithm based on high precision mobile platform is proposed.The direction vectors between the two feature corners in the overlapping field of vision before and after moving the conveyor belt are obtained by calculation,which are decomposed into unit components in the three coordinate axes,and the current Mosaic matrix can be obtained by combining them with the displacement values of the conveyor belt.Calculate the Euclidean distance between the feature corners after stitching and measure the stitching error.Two point clouds collected by the structured light camera before and after the moving conveyor belt were stitched to obtain the complete point cloud.Then,a joint conditional filtering and plane segmentation method for background plane separation is proposed.To solve the problem of point density in point cloud,a down-sampling method based on voxel filtering is introduced to simplify point cloud and facilitate subsequent data calculation and processing.According to the outlier noise in point cloud,the statistical filtering algorithm with more stable performance was selected to remove outlier and complete the pretreatment process to obtain the workpiece point cloud by comparing the characteristics of statistical filtering algorithm and radius filtering algorithm.Finally,several point cloud registration algorithms commonly used in testing are analyzed.According to the characteristics of the I-steel workpiece in this project,a rough registration method based on the intersection point and boundary contour of the projection point cloud boundary is proposed.The boundary line was obtained by fitting the projected plane point cloud boundary points,and the intersection points of the boundary line were aligned with the corner points of the model section to complete a registration.The boundary contour point cloud and model projection point cloud after the first registration were quickly ICP to complete the second registration.The point cloud of the workpiece after two registration provides a good initial position,which enables the subsequent ICP precision registration process to obtain high-precision registration results in a short time and complete the pose estimation of the workpiece.The robotic arm was used to measure the error distance of the test point,and the error of pose estimation was less than 6mm.