基于机器视觉的可回收垃圾智能分拣系统设计

Design of Intelligent Sorting System for Recyclable Waste Based on Machine Vision

以优傲机器人UR10与深度相机Intel RealSense D435为基础,设计了一种基于机器视觉的可回收垃圾智能分拣系统。系统采用张正友标定法对深度相机进行高精度标定,保证了机器人识别抓取的定位精度;采用基于深度学习的YOLO v4目标检测算法和KCF目标跟踪算法,实现了对可回收垃圾的姿态估计与精准识别抓取;基于ROS系统,实现了智能分拣系统的模块化,完成了智能识别分拣系统的整体构建。以常见可回收垃圾为实验对象对系统可靠性与稳定性进行了验证。结果表明:机器人对目标的抓取成功率达97.6%,分拣成功率为96.4%。因此,本系统能够实现对常见可回收垃圾的高精度识别定位及抓取,分拣成功率高、自动化程度高,可应用于实际生产场景。

An intelligent sorting system of recyclable waste is designed by using a UR10 robot and an Intel RealSense D435 depth camera.The system adopts Zhang Zhengyou calibration method to calibrate the depth camera with high precision,which ensures the positioning accuracy of the robot.YOLO v4 target detection algorithm and KCF target tracking algorithm based on deep learning are adopted to realize attitude estimation and accurate identification and grasping of recyclable garbage.Based on ROS system,the modularization of intelligent sorting system is realized,and the whole construction of intelligent sorting system is completed.The success rate and stability of the system are verified by taking common recyclable garbage as the experimental object.The results show that the success rate of target grasping is 97.6% and the success rate of sorting is 96.4%.Therefore,the system can realize high-precision identification,positioning and capture of common recyclable garbage,with a high sorting success rate and high degree of automation,which can be applied to actual production scenarios.