lamp-state-detection
收藏资源简介:
Lamp State Detection Dataset 是一个用于街道路灯灯泡检测的手动标注计算机视觉数据集。该数据集专注于灯泡本身(而非整个灯柱),旨在支持灯泡状态识别和下游地理感知基础设施监控的研究与应用。数据集包含1206张分辨率为1280×720像素的JPG图像,每张图像均标注了可见灯泡的边界框,并根据其状态标记为三类:lamp_on(灯泡亮起)、lamp_off(灯泡熄灭)和lamp_occluded(灯泡被遮挡)。数据以多种格式提供:每张图像对应一个JSON标注文件和一个YOLO格式的TXT标签文件。此外,还包含一个CSV格式的元数据文件,记录了每张图像的相对路径、文件名、视频ID、帧ID、时间戳(秒)、地理坐标(纬度、经度)、位置来源、定位方法、存在的标签类别、对象数量、边界框摘要以及图像尺寸等信息。数据收集于莫斯科的三条主要有轨电车线路沿线。该数据集被设计为更广泛应用管道的第一阶段,用于检测灯泡并分类其可见状态,关联检测结果与时间和GPS等元数据,以支持未来故障路灯的定位、制图和三角测量工作流程。与大多数关注整个灯柱检测的公开数据集不同,本数据集的核心区别在于其细粒度的目标——灯泡/发光区域。
The Lamp State Detection Dataset is a manually annotated computer vision dataset for street lamp bulb detection. Focusing on the bulbs themselves rather than entire lamp posts, it is designed to support research and applications for bulb status recognition and downstream geospatial infrastructure monitoring. The dataset comprises 1,206 JPG images with a resolution of 1280 × 720 pixels. Each image is annotated with bounding boxes for visible bulbs, which are categorized into three classes based on their operational status: lamp_on (illuminated), lamp_off (unlit), and lamp_occluded (partially or fully occluded). The dataset is provided in multiple formats: each image has a corresponding JSON annotation file and a TXT label file in YOLO format. Additionally, a CSV-format metadata file is included, which records detailed information for each image including its relative path, file name, video ID, frame ID, timestamp (in seconds), geographic coordinates (latitude and longitude), location source, positioning method, present label categories, number of objects, bounding box summary, and image dimensions. Data was collected along three major tram lines in Moscow. This dataset is intended as the first stage of a broader application pipeline for bulb detection and visible status classification, where detection results are linked with metadata such as timestamps and GPS to support future workflows for localization, mapping, and triangulation of faulty street lamps. Unlike most publicly available datasets that focus on detecting entire lamp posts, the core differentiator of this dataset is its fine-grained target: the individual bulb or light-emitting area.
数据集概述
数据集名称:Lamp State Detection Dataset
许可证:CC-BY-4.0
任务类型:目标检测(Object Detection)
语言:英语
数据规模:1,000 < 样本数 < 10,000
图像尺寸:1280 × 720 像素
数据集内容
该数据集包含 1206 张 手动标注的 JPG 图像,每张图像对应以下文件:
- JSON 标注文件(
annotations/) - YOLO 格式 TXT 标签文件(
labels/) - 元数据 CSV 文件(
metadata.csv)
标签类别(共3类):
lamp_on:灯泡可见且点亮lamp_off:灯泡可见但未点亮lamp_occluded:灯泡区域部分遮挡或视觉模糊
目录结构:
lamp-state-detection/ ├── images/ ├── annotations/ ├── labels/ ├── metadata.csv ├── data.yaml └── README.md
元数据字段
每个样本的元数据包含以下字段:
relative_pathfile_namevideo_idframe_idtimestamp_seclatitudelongitudelocation_sourcelocation_methodlabels_presentnum_objectsbboxes_summaryimage_widthimage_height
数据采集
数据采集于 莫斯科,覆盖三条主要电车路线:
- 从 Kuskovo 地区驶向市中心的公路路线
- 围绕莫斯科城的路线,始于 Kutuzovsky Ave,沿 A106 公路向 Rublyovo-Uspenskoye 延伸
- Ramenki 路线,向 Aminevskoye / Rublevskoye 方向
设计初衷与应用
该数据集聚焦于 路灯灯泡本身(而非灯杆或灯具整体),旨在支持:
- 灯泡状态检测与识别
- 结合时间、GPS 等元数据的关联分析
- 未来应用于故障路灯的 定位、地图构建与三角测量
局限性
当前版本为 2D 检测数据集,尚未定义多视角对应关系、相机姿态估计或实例级3D地面真值坐标,因此不适合直接用于三角测量任务。




