铁道科学与工程学报2025，Vol.22Issue(9)：4145-4159,15.DOI:10.19713/j.cnki.43-1423/u.T20241933
驼峰编组站摘钩机械臂视觉检测系统研究
Visual detection system of uncoupling manipulator in hump marshalling yards
摘要
Abstract
In view of short time requirement for automatic uncoupling operations in hump marshalling yards,and difficulty in quickly and accurately detecting,identifying,and locating the vehicle's hook rod handles and foot pedals in complex environments,a visual detection and target localization method based on an active RGBD camera combined with an improved YOLO11 deep learning algorithm was proposed.Firstly,according to the automatic uncoupling principle of the manipulator and its application requirements,the visual detection system scheme design and RGBD camera selection were carried out.Using the HGNetV2 lightweight network(GC-HGNetV2)improved based on GhostConv as the backbone network of YOLO11 can enhance its feature extraction capability while ensuring model lightweighting.At the same time,by introducing the EMASlideLoss loss function,the problem of imbalanced positive and negative samples was effectively solved.The stability of the model training process was improved.The detection performance of the model was further enhanced.The experimental results on the dataset of hook rod handles and foot pedals show that compared with the original YOLO11 model,the improved algorithm has increased the R and mAP@50 metrics by 0.3 percentage points and 0.6 percentage points respectively.It can effectively solve the problem of false detection and missed detection caused by the single characteristics of hook rod handles and foot pedals in complex environments,resulting in higher detection accuracy and robustness.The automatic uncoupling manipulator and visual detection system are designed and constructed in the laboratory,and the hand-eye calibration and the automatic uncoupling test of the manipulator are conducted.The experimental results show that the designed visual detection system can quickly and accurately detect and identify hook rod handles and foot pedals in complex backgrounds,and accurately determine their three-dimensional coordinates in the coordinate system of the manipulator,assisting the manipulator in completing the automatic uncoupling tasks.It has good real-time response performance and can meet the needs of automatic uncoupling manipulator applications.关键词
自动摘钩机械臂/主动式RGBD相机/YOLO11/GC-HGNetV2/EMASlideLossKey words
automatic uncoupling manipulator/active RGBD camera/YOLO11/GC-HGNetV2/EMASlideLoss分类
交通工程引用本文复制引用
罗涪文,王雪梅,倪文波,韦蕾蕾,艾明阳..驼峰编组站摘钩机械臂视觉检测系统研究[J].铁道科学与工程学报,2025,22(9):4145-4159,15.基金项目
四川省国际科技创新合作项目(2022YFH0038) （2022YFH0038）
