Research On Point Cloud Segmentation And Pose Estimation Of Mixed-loaded Pallet Depalletizing Based On Three-dimensional Vision

In recent years,with the rapid development of e-commerce industry,the supporting warehousing and logistics industry also develops rapidly.The warehousing and logistics industry has the characteristics of a wide variety of goods and a large number of goods.In order to improve the utilization of space and reduce transportation and storage costs,mixed palletizing gradually replaces plastic palletizing and plays an important role in the warehousing and logistics industry.In the field of mixed palletizing and disassembling,due to the unknown size of the box,there are problems such as height difference,etc.,the traditional two-dimensional vision technology is difficult to meet the requirements of positioning accuracy,and the mainstream three-dimensional vision technology has the problems of large time consumption and insufficient stability.In this paper,combined with three-dimensional vision technology,point cloud segmentation and pose estimation algorithm for mixed palletizing and disassembling are proposed,and an experimental platform is built to study its positioning accuracy,stability and real-time performance.According to the requirements of palletizing system,the overall scheme and working flow of hybrid palletizing system are determined.Aiming at the recognition requirements of mixed stack with fast imaging speed,large recognition area and low precision demand,a TOF technology based snapshot depth camera was selected.Then the principle and algorithm of system calibration are stud ied,including internal reference calibration of depth camera,TCP calibration of robot and hand-eye calibration of system.The pre-processing algorithm of mixed stack point cloud is studied.Point cloud filtering,point cloud simplification and point cloud smoothing are carried out to solve the problems of noise point cloud,point cloud void and large amount of data in the original point cloud data.The algorithm of hybrid stack point cloud segmentation and pose estimation is studied.According to the characteristics of the height difference of mixed palletizing,a two-step segmentation strategy is proposed in this paper.In the first step,a linear model of RANSAC optimal threshold and height difference between mixed palletizing planes is established to fit and extract the point cloud plane of mixed palletizing.The second step is to segment and cluster the point cloud on the box surface by preprocessing and region growing algorithm.Then,the segmented point cloud plane is used to study the pose estimation algorithm according to its spatial position information.Finally,the pose information of the plane to be captured is output according to the order of priority.Finally,the system implementation and experimental verification are carried out.Firstly,the hardware selection and construction of the experimental equipment are completed,and then the software framework of the palletizing sy stem is built based on the ROS system.The system calibration was completed through the experiment and the calibration error was analyzed.Finally,the accuracy,stability and real-time performance of the palletizing system were measured through the experiment,and the final experimental results met the requirements of palletizing.