Research On Object Localization And Sorting Algorithm For Logistics Environment

With the continuous development of society and economy,the scale of the logistics industry has become larger and larger,and labor costs and the degree of automation have become factors that limit the faster development of the logistics industry.High and new technology represented by robotics is also difficult to promote the automation level of the logistics industry because of the complex logistics scenarios.Starting from a typical logistics scenario,this article addresses the difficulties in the orderly sorting problem in the case of disorderly and disorderly stacking of items in this scenario,and decomposes the entire orderly sorting process into the positioning problem of a single object in the disorderly and disorderly stacking situation and The best grasping or drawing pose of a single object generates a problem.Through the use of robotics,image processing,deep learning and other technologies,this paper designs corresponding algorithms to solve the above problems,verifies the effectiveness of the algorithm through experiments in real scenarios,and gives a complete set of vision-based Robotic arm automatic sorting system.Aiming at the positioning problem of a single envelope and package in a disorderly stacked environment,this paper adopts a deep learning-based method and uses an RGBD camera.In a laboratory environment,the situation of mutual occlusion and stacking between objects was simulated,and a data set was made by using Real Sense camera to collect images.For the location of the envelope,this paper uses a single-stage detection method based on RGB images to obtain the bounding box information of a single envelope.For the positioning of the package,considering the deformation characteristics of the package and the requirements of grasping accuracy,the method of instance segmentation based on RGBD image is used to obtain the contour information of a single package.The result of the single object positioning problem in the case of a disorderly pile is given by the envelope bounding box and the contour information of the package.After determining the location information of a single logistics item,this paper also needs to solve the problem of generating the best picking or grasping pose of a single item.For envelopes and packages,although different sorting methods are used in this paper,the pose generation algorithm of the same process is designed.The algorithm is based on the sampling evaluation process.First,it searches for candidates that meet the restriction conditions to acquire suction or grasp poses in the object area given by the localization results,and then scores each candidate pose through an evaluation network,and finally selects the highest score one.For different sorting methods,the only thing need to done is to set different restriction conditions in order to find out the candidate poses that meet the conditions.Finally,a complete set of vision-based robotic arm hardware system was built in the laboratory environment and the software system design was completed under the open source robot operating system ROS.Then simulated the real logistics scene and carried out the actual robotic arm sorting experiment.Experiments show that the designed algorithm can solve the sorting of typical logistics items,and the whole set of vision-based robotic arm automatic sorting system is feasible.