| There are a large number of material handling tasks in the industrial production field,and pallets are the most commonly used material carriers.Automatic guided vehicles(AGV)can achieve efficient material handling by picking up,transporting,and placing pallets.In the process of picking up the pallet,the pallet is usually fixed at a certain position,and the industrial robot is guided to travel to the fixed pallet position through external magnetic strips,track assistance,or scene mapping and positioning technology to complete the pickup.However,blindly picking up the pallet without identifying and positioning it will lead to a decrease in efficiency and safety hazards.Therefore,pallet recognition and positioning have research significance.Existing pallet recognition and positioning methods mainly use neural networks and deep learning technology,which have higher complexity,poor generalization for different styles of pallets,and insufficient positioning accuracy.To address these problems,this thesis proposes a pallet recognition and positioning system based on a depth camera.First,this thesis surveys the existing pallet recognition and positioning methods at home and abroad,and compares and analyzes various sensors and different algorithms,points out the existing problems,and introduces the work content and article structure of this thesis.Second,in response to the existing problems,this thesis innovates on the algorithm and proposes a pallet recognition method based on pallet pillar feature matching.Based on the template matching idea,the pallet pillars are extracted by point cloud filtering,segmentation,and clustering,and the pallet is recognized by combining the pallet’s geometric features.A lightweight and robust pallet recognition algorithm is realized.At the same time,to address the problem of inadequate positioning accuracy,a high-precision pallet pose estimation method is designed.The prior pose obtained from the first frame is quickly processed in subsequent data,and multiple frames of data are selectively feature-fused to reduce single-frame data fluctuations and enrich the pallet pillar point cloud,improving the pose calculation accuracy.Based on the above pallet recognition and positioning algorithm,a complete pallet recognition and positioning system is designed,including the design of hardware boards and the development of software functions.The entire embedded platform is deployed directly on the industrial robot as an independent module,which identifies and locates the pallet,outputs its pose,generates a path,and guides the AGV for automatic pallet pickup and transportation.Finally,the algorithm performance is analyzed and tested.The entire system is deployed on an industrial robot and undergoes fully automated logistics testing in a real factory environment.1000 data points are collected,and the results show that the pallet recognition accuracy reaches 99%,and the positioning accuracy reaches 10 mm and 0.5°.Accurate pallet picking can be achieved in complex factory environments. |
