车身表面缺陷单目定位数据
收藏资源简介：
本数据集面向轨道交通领域机器人视觉检测与缺陷定位研究，主要用于高铁车身表面缺陷单目定位精度验证、算法性能评估及系统优化，旨在为工业机器人视觉引导与漆膜缺陷定位提供标准化、可复用的数据支撑。数据依托国家重点研发计划相关课题，由武汉理工大学汽车工程学院硕博研究生依据国家标准GB/T 45501-2025《工业机器人三维视觉引导系统通用技术要求》在武汉理工大学汽车工程学院完成，数据采集与整理工作于2023年7月15日进行，总规模为169MB。在数据采集与处理过程中，采用ABB IBR7600工业机器人搭载单目海康工业相机（型号 MV-CS050-10UC）及MVL HF1228M 6MP工业镜头进行标定与定位验证。首先，通过对标定板进行65张图像采集并完成手眼标定，建立相机像素坐标与机器人末端坐标系之间的转换关系。随后，对车身表面进行图像采集，使用YOLOv5漆膜缺陷检测算法提取缺陷点，并基于像素点逆投影方法计算缺陷点三维坐标。为验证定位精度，在车身表面粘贴标记点，利用相机及相同逆投影方法获取计算三维坐标，同时通过机器人末端探针进行示教测量获取真实三维坐标。将计算值与实测值进行对比，统计X、Y、Z方向坐标偏差，并计算平均误差，作为单目定位精度评价指标。本数据集内容包括手眼标定图片（标定板拍摄图像）、验证图片（标记点拍摄图像）、缺陷定位图片（通过YOLOv5检测后的缺陷图像）及定位数据表。本数据记录了手眼标定结果、定位精度数据及缺陷点三维坐标，系统呈现了缺陷定位全过程及精度分析结果。数据完整反映了单目视觉在车身缺陷定位中的测量精度、像素坐标与三维坐标转换关系及误差统计，为后续缺陷检测算法优化、定位方法改进和机器人视觉系统性能评估提供了可靠的数据基础。
This dataset is targeted at the research of robotic visual inspection and defect localization in the rail transit field, and is mainly used for monocular positioning accuracy verification, algorithm performance evaluation and system optimization of high-speed train body surface defects. It aims to provide standardized and reusable data support for industrial robot visual guidance and paint film defect localization. The dataset was developed under the relevant projects of the National Key R&D Program of China, and was completed by master's and doctoral graduate students from the School of Automotive Engineering, Wuhan University of Technology in accordance with the national standard GB/T 45501-2025 *General Technical Requirements for Industrial Robot 3D Visual Guidance Systems* at the School of Automotive Engineering, Wuhan University of Technology. Data collection and organization were carried out on July 15, 2023, with a total size of 169 MB. During the data collection and processing process, an ABB IBR7600 industrial robot equipped with a monocular Hikvision industrial camera (model MV-CS050-10UC) and an MVL HF1228M 6MP industrial lens was used for calibration and positioning verification. First, 65 images of the calibration board were collected and hand-eye calibration was completed to establish the conversion relationship between the camera pixel coordinates and the robot end-effector coordinate system. Subsequently, image collection was performed on the train body surface. The YOLOv5 paint film defect detection algorithm was used to extract defect points, and the three-dimensional coordinates of the defect points were calculated based on the pixel inverse projection method. To verify the positioning accuracy, marker points were pasted on the train body surface. The calculated three-dimensional coordinates were obtained using the camera and the same inverse projection method, while the ground-truth three-dimensional coordinates were obtained through teaching measurement using the robot end-effector probe. The calculated values were compared with the measured values, the coordinate deviations in the X, Y, and Z directions were counted, and the average error was calculated as the monocular positioning accuracy evaluation metric. The contents of this dataset include hand-eye calibration images (images captured of the calibration board), validation images (images captured of the marker points), defect localization images (defect images detected by YOLOv5), and localization data tables. This dataset records the hand-eye calibration results, positioning accuracy data, and three-dimensional coordinates of defect points, and systematically presents the entire process of defect localization and the accuracy analysis results. The dataset fully reflects the measurement accuracy of monocular vision in train body defect localization, the conversion relationship between pixel coordinates and three-dimensional coordinates, and error statistics, providing a reliable data foundation for subsequent defect detection algorithm optimization, localization method improvement, and industrial robot vision system performance evaluation.
