轨道异物检测可见光图像标注数据

本数据集是一个面向铁路周界入侵检测的可见光图像标注数据集，包含可见光图像数据和其对应的标签。数据集由4个文件夹组成，每个文件夹下有两个文件夹分别记录了可见光图像数据和其对应的标签。数据集覆盖了多种典型入侵场景，并兼顾白天、夜晚、晴天、雨天等不同光照和天气条件，充分体现了复杂铁路环境下入侵事件的多样性和随机性。该数据集在数据采集、整理及标注方面进行了严格的质量控制，确保数据的真实性、完整性与可用性。该数据集为行人入侵轨道检测领域提供了宝贵的可见光图像标注资源，对于推动铁路周界行人入侵检测等研究任务具有重大潜力。通过公开该数据集，有助于基于可见光图像的行人入侵检测算法的研究发展，为科研人员提供宝贵的数据资源。通过公开共享，可显著推动铁路周界入侵检测技术的研究进展，并为提升轨道交通运行安全和智能化水平提供重要的数据基础。
在雄安R1线高架段布设地面监测基站及试验车上安装实验平台，配备两个森云GMSL相机（分辨率1920×1080），传感器的频率均为10HZ，所有传感器通过GNSSPPS信号与硬件触发模块实现微秒级时间同步，空间坐标经外参标定后统一至本地东北天（ENU）坐标系。图像由运行于 Ubuntu20.04+ROSNoetic 的主控计算机采集，主控机器为爱视图灵CES-RUG-ORIN-701，使用 usb_cam 驱动节点发布话题，并通过rosbagrecord 命令将话题同步入 .bag 文件。解包后，得到异物检测可见光图像，使用labelme标注软件标注数据，用于后续模型训练，针对雄安新区城市轨道交通规划和线路环境特点，训练得到异物检测模型，开展多源融合的路网净空安全智能感知体系与装备研究，实现对线路障碍物、周界入侵、自然灾害等威胁行车安全事件的全面感知，为保障雄安新区轨道交通的安全运营提供支撑。数据量10.72GB。

This dataset is a visible light image annotation dataset for railway perimeter intrusion detection, containing visible light image data and their corresponding labels. The dataset consists of 4 folders, each of which contains two sub-folders storing the visible light image data and their corresponding labels respectively. The dataset covers multiple typical intrusion scenarios, and takes into account different lighting and weather conditions such as daytime, nighttime, sunny days, and rainy days, fully reflecting the diversity and randomness of intrusion events in complex railway environments. Strict quality control has been implemented in data collection, organization and annotation to ensure the authenticity, integrity and availability of the data. This dataset provides valuable visible light image annotation resources for the field of pedestrian track intrusion detection, and has great potential for promoting research tasks such as railway perimeter pedestrian intrusion detection. Publishing this dataset will facilitate the research and development of pedestrian intrusion detection algorithms based on visible light images, providing precious data resources for researchers. Public sharing can significantly promote the research progress of railway perimeter intrusion detection technology, and provide an important data basis for improving the operational safety and intelligence level of rail transit.

Ground monitoring base stations were deployed on the elevated section of Xiong'an Rail Line R1, and an experimental platform was installed on the test train, which was equipped with two Senyun GMSL cameras with a resolution of 1920×1080. Both sensors operated at a frequency of 10 Hz. All sensors achieved microsecond-level time synchronization via the GNSS PPS signal and hardware trigger module, and their spatial coordinates were unified to the local East-North-Up (ENU) coordinate system after external parameter calibration. The images were collected by a main control computer running Ubuntu 20.04 and ROS Noetic, where the main control unit was Aiview CES-RUG-ORIN-701. The usb_cam driver node was used to publish ROS topics, and the topics were synchronized into .bag files via the rosbag record command. After decompressing the collected .bag files, visible light images for foreign object detection were extracted, and the data was annotated using the LabelMe annotation software for subsequent model training. Aiming at the characteristics of urban rail transit planning and line environment in Xiong'an New Area, a foreign object detection model was trained, and research on the multi-source fusion-based intelligent perception system and equipment for road network clearance safety was conducted, enabling comprehensive perception of traffic safety threats including line obstacles, perimeter intrusion, natural disasters and other hazards, so as to provide support for the safe operation of rail transit in Xiong'an New Area. The total data volume of the dataset is 10.72 GB.

搜集汇总

背景与挑战

背景概述

该数据集是一个面向铁路周界入侵检测的可见光图像标注数据集，包含多种典型入侵场景的图像和标签，覆盖不同光照和天气条件，数据量10.72GB，标注严格，适用于行人入侵检测算法的研究。