Design Of A 3D Point Cloud-based Intelligent Sorting System For Robots

The development of information technology and intelligent control technology is driving the continuous improvement of intelligent manufacturing.Among them,sensing technologies in smart manufacturing,including target detection,rapid positioning,and precision measurement,play an important role and are widely used in various aspects of processing and manufacturing.In this thesis,we take robot intelligent sorting and measurement system as the research object,combine deep learning technology with recognition and positioning requirements to enhance robot intelligent sorting capability;meanwhile,we build an integrated robot sorting system and structured light measurement system to complete rapid intelligent recognition of target objects and 3D parameter detection system.The main contents of this thesis include:Firstly,in order to improve the level of sorting intelligence and positioning accuracy,based on the application of YOLOv5 s for object recognition and positioning,a depth image center positioning method based on corner point detection is proposed so as to solve the problem that the recognition positioning point deviates from the center of the upper surface of the object.The center pixel coordinates are aligned with the corresponding points of the depth map to achieve accurate object center positioning.Secondly,in order to improve the robustness and accuracy of the structured light 3D reconstruction system,a 3D point cloud reconstruction method based on local raster complementation is proposed for the missing reconstruction problem caused by the reflective characteristics of the surface of metal objects.The optimal threshold is calculated by normalizing the raster image pixels and binarizing the captured raster map based on the threshold;the ROI area is generated according to the missing position of the raster map;the complete raster stripes are selected and filled into the ROI area to generate the complete raster.Finally,the experimental platform of robot intelligent sorting system based on 3D point cloud is constructed,and the hand-eye calibration of the system is completed.Based on the improved center positioning method,the target object is identified and positioned by the camera to achieve sorting,and the robotic arm is guided to grasp the object;based on the improved raster complementation method,the target object is reconstructed in 3D,and the data related to its point cloud model is detected and compared.The experiments show that the improved positioning algorithm can realize the positioning of the center of the upper surface of the object,and the pixel distance from the reference center point is stable within 8pixels;the improved reconstruction algorithm can effectively eliminate the influence of reflections on the surface of the object and improve the reconstruction integrity;the overall sorting system improves the detection and sorting accuracy of the object based on the improved method.