Research On Scatered Parts Identification And Localization Based On Point Cloud Processing

As an important part of in the field of industrial production,assembly has high intensity and repeatability,and has a certain risk in certain environment.This process has mainly depended on manual sorting for many years.It has high production cost,low accuracy,and slow speed.The industrial robot is applied in the field of assembly,the automatic grasping and assembling of parts can be realized,and the intelligent degree of industrial productivity and automatic assembly could be greatly improved.The difficulty of automatic assembling is target detection.The random bin picking technology based on 3D vision has gradually become an effective method to solve this problem because of its huge application potential.It is also a hot topic for domestic and foreign scholars.This paper proposes a method for recognition and positioning scattered parts based on 3D point cloud processing.The 3D vision measurement system is used to obtain the point cloud data on the surface of the scattered parts in the box.The 3D point cloud processing algorithm is used to obtain the pose information of the parts.The subject has deeply studied two key issues of scene segmentation and registration in the process of identification and localization.Aiming at the difficulty of scene segmentation in the process of robot random bin picking,a point cloud segmentation method based on improved Euclidean clustering is proposed.In the preprocessing stage,an iterative radius filtering method is proposed to remove outliers,and the parts cloud after removing the interference point is obtained;The edge detection method based on normal angle removes edge points in the point cloud,separates the workpiece that collides with each other,and then uses the improved,search radius adaptive Euclidean clustering method to carry on the point cloud segmentation,many workpiece point cloud subsets are obtained.Finally,completing the point cloud segmentation according to the method based on distance constraint.In addition,the offline registration of template cloud point provides the basis of parameter selection for segmentation,which ensures the accuracy of segmentation results and improves the segmentation speed.The experimental results show that the method can accurately segment the workpiece of interest,and the segmentation speed is about 635 ms,which satisfies the real-time requirement of industrial robot crawling.For the problem of parts identification and localization,a point cloud registration method based on the fusion of SHOT features is proposed.Firstly,the template point cloud feature database is established offline to be used for the subsequent pose calculation of individual parts.Then,the position and pose estimation of each part point cloud subset obtained by segmentation is performed online,and a method based on the direction bounding box clipping is proposed to get the part point cloud after removing the sparse edge point,and to obtain the key point set with the uniform sampling algorithm;The key points are uniquely described by improving SHOT feature descriptors;and the key points of part point cloud are found in template point cloud by using minimum variance method,and the initial transformation matrix is solved according to the corresponding relation;Finally,the accurate alignment of the part is obtained by using the ICP algorithm.In order to verify the effect of the position and attitude estimation algorithm,the hardware platform is built by the projector and the industrial camera,and the point cloud data of parts in different scattered scenes are obtained.The pre-processing,segmentation and registration of the collected scene cloud data were used to obtain the pose information of the parts.The experimental results were statistically analyzed.The results show that the algorithm can accurately and quickly segment the parts and the registration results.Compared with the SHOT feature registration and the accuracy of the FPFH feature registration algorithm,the accuracy was increased by 37.10% and 30.07%,respectively,and the registration speeds were increased by 21.21% and 35.64% respectively.