An Intelligent Robot Sorting System By Deep Learning On RGB-D Image

Industrial robots have seen wide adoption in the field of intelligent manufacturing, and they are applied to the intelligent sorting of items on the industrial product line. However, such techniques have some limitations, such as the object being placed at a fixed height. To solve this problem and meet the needs of flexible manufacturing, this paper develops an intelligent robot sorting system by deep learning on RGBD image (RSSDR). RSSDR consists of five modules, including central host, robot, visual kit, microservices and machine visual control software. The central host provides the image of 3D printed industrial parts collected by the visual kit, and calls the deep learning algorithm on the microservice to detect the key points on the image, and then sends the position of the grasping points to the robot through the machine vision control software to control the robot to sort the 3D printed industrial parts. In RSSDR, to visually identify a total of 3 different 3D printed industrial parts in uncontrolled conditions, YOLOv5s is used as the detection model, which achieves 97.65% accuracy and 35ms runtime. The experimental results show that RSSDR is suitable for sorting electronic component.