有限空间作业不安全行为视觉数据集

该数据集包含多场地、多角度、多距离、光照各异的1000余条视频数据，以及从中抽取的13万多帧图像数据。其中，每条视频均具有对应的操作流程、作业人员属性、现场设备属性标注；每帧图像均具有目标框、4种帧属性、6种人体属性、现场设备属性等标注，标注的目标实例框总量达到220多万个。应用场景：该数据集针对有限空间作业事故率高、监管难的痛点，为训练高精度、强鲁棒性的有限空间安全监控模型提供了坚实的数据基础，推动了算法研究的深入，能有效降低复杂真实环境下的误报漏报，并支持识别复杂组合风险，有助于降低事故率，提升行业安全认知水平。另外，该数据集将赋能有限空间作业安全的智能化升级，推动事故预防与安全生产水平提升。基于其开发的AI技术可构建实时监控与预警系统，在作业现场即时识别防护缺失、违规操作等高风险行为并报警。同时，该数据集可以支撑自动化安全审计与人员培训，通过视频分析生成合规报告、追溯事故根源，通过历史数据分析预测潜在风险点，指导优化作业规程和安全资源配置。

This dataset contains over 1,000 video recordings with diverse locations, multiple shooting angles, varying distances and different lighting conditions, as well as more than 130,000 image frames extracted from these videos. Each video is annotated with corresponding operation procedures, attributes of on-site operators and attributes of on-site equipment. Each image frame is annotated with target bounding boxes, 4 types of frame attributes, 6 types of human body attributes, on-site equipment attributes and other relevant information, with the total number of annotated target instance bounding boxes exceeding 2.2 million.

Application scenarios: Aiming at the pain points of high accident rate and difficult supervision in confined space operations, this dataset provides a solid data foundation for training high-precision and highly robust confined space safety monitoring models, promoting in-depth progress in algorithm research. It can effectively reduce false positives and false negatives in complex real-world environments, support the recognition of complex combined risks, help lower accident rates and enhance the industry's safety awareness level. In addition, this dataset will empower the intelligent upgrading of confined space operation safety, promoting accident prevention and work safety improvement. The AI technology developed based on this dataset can build a real-time monitoring and early warning system, which can instantly identify high-risk behaviors such as lack of personal protective equipment and illegal operations at the operation site and trigger alarms. Meanwhile, this dataset can support automated safety audits and personnel training: generate compliance reports and trace the root causes of accidents through video analysis, predict potential risk points via historical data analysis, and guide the optimization of operation procedures and safety resource allocation.