地铁轨道现场道床异物巡检数据

地铁轨道道床智能巡检系统专注于异物检测与安全预警，主要应用于：通过工电隧综合巡检机器人采集的地铁道床图像，结合AI算法分析并识别道床上榔头、螺丝钉组合盒、绝缘手套、螺丝刀等异物，判断道床内是否存在异物等异常状态，实现轨道道床的状态预警与报警，为地铁运营提供智能化安全保障。本系统通过搭载于综合巡检机器人上的高分辨率线扫相机，在列车以15km/h运行过程中实时采集轨道道床区域图像，结合YOLOv8目标检测算法实现对道床异物的自动识别与预警。
1.数据采集阶段：根据行车方向和采样频率设置相机参数，确保图像连续、完整覆盖轨道两侧区域。
2.数据预处理：采集到的图像经去雾、增强、去噪等预处理操作后送入目标检测模型进行分析。
3.YOLOv8检测模型经过大量标注数据及模型训练，可识别榔头、螺丝钉组合盒、绝缘手套、螺丝刀等多种常见道床异物，输出其位置框、类别及置信度。
4.系统依据异物尺寸、位置风险等级进行分级判断，当识别状态为：有异物时，以“1”（预警）形式上传至可视化平台。当识别状态为：正常时，以“0”（正常）实现实时报警功能，便于现场人员快速判别设备的状态是否异常；
5.系统支持按日、月、年等时间维度统计异物分布情况，生成趋势分析报表，辅助运维人员制定清理策略，提升轨道运行安全性与智能化管理水平。

The intelligent inspection system for subway ballast bed focuses on foreign object detection and safety early warning. Its core application scenarios are as follows: collecting subway ballast bed images via the integrated inspection robot for subway track works, electrical systems and tunnel maintenance, combining with AI algorithms to analyze and identify foreign objects such as hammers, screw kits, insulated gloves and screwdrivers on the ballast bed, judging whether there are abnormal conditions including foreign objects in the ballast bed, realizing status early warning and alarm for the track ballast bed, and providing intelligent safety guarantees for subway operation.
This system uses a high-resolution line-scan camera mounted on the integrated inspection robot to collect images of the ballast bed area in real time when the train is running at 15 km/h, and combines the YOLOv8 object detection algorithm to realize automatic identification and early warning of track bed foreign objects.
1. Data collection stage: Set camera parameters according to the driving direction and sampling frequency to ensure that the images continuously and completely cover the areas on both sides of the track.
2. Data preprocessing: The collected images are subjected to preprocessing operations such as defogging, enhancement and denoising, and then sent to the object detection model for analysis.
3. The YOLOv8 detection model, trained on a large amount of annotated data, can recognize a variety of common ballast bed foreign objects such as hammers, screw kits, insulated gloves and screwdrivers, and output their bounding boxes, categories and confidence scores.
4. The system conducts hierarchical judgment based on the size and position risk level of foreign objects. When foreign objects are detected, it uploads "1" (early warning) to the visualization platform. When the status is normal, it uploads "0" (normal) to realize the real-time alarm function, enabling on-site personnel to quickly judge whether the equipment status is abnormal;
5. The system supports counting the distribution of foreign objects in time dimensions such as day, month and year, generating trend analysis reports, assisting operation and maintenance personnel in formulating cleaning strategies, and improving track operation safety and intelligent management level.