3D Reconstruction Of Spraying Piece Based On RGB-D Camera

With the promotion of the concept of "Industry 4.0",the spraying assembly line has gradually developed to the automatic mode.Domestic small and medium-sized spraying enterprises spray more types of workpiece,small batch,its spraying method is mostly manual spraying,not only high labor cost,and toxic spray painting damage to the human body.In order to replace manual spraying,adapt to the characteristics of small batch of small and medium-sized spraying enterprises,spraying robot came into being,but spraying robot programming is still difficult.In the study of spraying robot,path planning and robot autonomous programming are both based on the artifact 3D model.In view of the difficulty of difficult to obtain complex 3D model by using lowcost imaging scheme,this paper proposes the online reconstruction of spraying line workpiece based on multiple consumer RGB-D cameras.This paper mainly completes the following work:(1)Design and calibration of 3D imaging system for workpiece in spraying production line.The imaging system designed in this paper is composed of four RGBD cameras and a simulation experiment platform.The cameras are placed in four directions of the simulation experiment platform.The workpiece is suspended on the platform slideway,and when it reaches the preset area,four cameras capture a frame of RGB-D image.Aiming at the problem of multi-view image fusion in opposite camera areas with no/less overlapping area,an extrinsic parameter calibration method based on RGB-D camera is proposed.In this paper,a special double-sided calibration plate with a triangular pattern is designed,and the rotation and translation matrix T is calculated according to the vertex(reference point)corresponding to the triangle as the same point in the world coordinate system.Triangle edges in the calibration plate are first extracted and converted into point clouds using multi-scale edge detection.Secondly,for the problem that the point cloud of the triangle edge has no intersection in space,use RANSAC to fit the straight line to obtain the straight line equation.The closest point between the three straight lines is taken as the reference point,and the centroid of the point set is moved to the origin,and the singular value decomposition is used to obtain the R matrix.The two RGB-D cameras in the no/less overlapping area are calibrated by the extrinsic parameters designed in this paper,and the point cloud in the camera coordinate system can be directly fused with the point cloud in the opposite camera coordinate system.(2)Fast optimization of depth maps based on RGB map guidance.Aiming at the problem that the point cloud of the 3D workpiece obtained directly from the depth map has many noises around and the edges are not smooth enough,this paper proposes a fast optimization algorithm for depth map based on RGB map guidance.The algorithm uses the Background Matting depth model to accurately segment the foreground of the workpiece in the RGB image,and uses the joint bilateral filtering guided by the foreground of the workpiece to filter the workpiece region of the depth map and segment the depth map workpiece.Using the high-quality characteristics of the RGB image,the gray domain weight of the corresponding area of the RGB image is introduced into the edge area of the workpiece depth image through the mapping relationship between the RGB image and the depth image.Due to the stability and accuracy of the grayscale domain,joint bilateral filtering is only used in the precise workpiece area,which reduces redundant computation and has the effect of edgepreserving and denoising for the workpiece area.Experiments show that this method can quickly optimize the depth map,and at the same time use it as a mask to segment the workpiece depth map,and finally obtain a high-quality workpiece point cloud.(3)3D reconstruction of spraying workpiece based on multi-visual fusion.In view of the long time consuming to convert all point clouds into the first camera coordinate system,this paper proposes the scheme of efficiently integrating point clouds under the four camera coordinate systems.The rotation translation matrix T of the RGB-D camera is obtained by the external parameter designed,which directly integrates the point cloud in the camera coordinate system with the point cloud in another opposite camera coordinate system,which is less time-consuming.Based on the point cloud under the less / no overlapping area camera twice,the 3 D reconstruction of the spraying piece can be completed with only one registration between adjacent cameras.In this paper,Super-4PCS is used for global registration,and fast ICP to improve registration speed.Based on the 3D reconstruction method of spraying pipeline workpiece developed by consumer RGB-D camera,the performance of the method is high quality and high efficiency,and the reconstruction accuracy is 8mm and time consuming 0.762 s,which can meet the requirements of online reconstruction.