电子科技2026，Vol.39Issue(10)：10-23,14.DOI:10.16180/j.cnki.issn1007-7820.2026.10.002
基于改进DETR-YOLO-Pose的托盘定位算法
Tray Localization Algorithm Based on Improved DETR-YOLO-Pose
摘要
Abstract
Pallet identification and positioning is a key challenge that unmanned forklifts face in complex warehouse environments.Problems such as pallet occlusion and stacking lead to low efficiency in identification and positioning.A monocular pallet positioning method combining target detection and key point detection is proposed.The REF-DETR(Reparametrized Focus-diffusion-Detection Transformer)algorithm is adopted for pallet detec-tion,which improves the recognition accuracy in various complex environments.The extraction accuracy of the pal-let contour is improved through key point detection and adaptive edge adjustment.For the estimation of pallet pose,the calculation methods of pallet inclination Angle and distance were proposed.The LMedS(Least Median of Squares)algorithm has improved the accuracy and stability of pose estimation.The experimental results show that the pallet recognition accuracy of the REF-DETR algorithm is 92.8%,and the FPS(Frames Per Second)is 108.1 frame·s-1.Compared with the original algorithm,the accuracy and FPS of the proposed algorithm have increased by 4.1 percentage points and 38.4 frame·s-1 respectively.The average error of the inclination angle is less than 3.2°,and the average error of the horizontal distance is less than 30 mm.It has good positioning performance and practi-cal value.关键词
REF-DETR/托盘定位/目标检测/关键点检测/自适应边缘调整/位姿估计Key words
REF-DETR/tray positioning/object detection/key point detection/adaptive edge adjustment/pose estimation分类
信息技术与安全科学引用本文复制引用
周著国,鲁玉军,吕利叶..基于改进DETR-YOLO-Pose的托盘定位算法[J].电子科技,2026,39(10):10-23,14.基金项目
浙江省重点研发项目(2022C01242) （2022C01242）
浙江理工大学龙港研究院项目(LGYJY2021004) Key R&D Program of Zhejiang(2022C01242) （LGYJY2021004）
Longgang Research Institute Project of Zhejiang Sci-Tech University(LGYJY2021004) （LGYJY2021004）
