基于改进YOLOv8n模型的轨道交通轨枕精确定位技术
闫晓夏1郭建钦1翟胡超2刘玉涛1李俊鑫1王亮先1
Precise Rail Transit Sleeper Positioning Technology Based on Improved YOLOv8n Model
YAN Xiaoxia1GUO Jianqin1ZHAI Huchao2LIU Yutao1LI Junxin1WANG Liangxian1
-
作者信息:1.铁科院（深圳）研究设计院有限公司， 518034， 深圳
2.广东城际铁路运营有限公司， 511458， 广州
-
Affiliation:1.Shenzhen Research and Design Institute of CARS, 518034, Shenzhen, China
2.Guangdong Intercity Railway Operation Co., Ltd., 511458, Guangzhou, China
-
关键词:
-
Key words:
-
DOI:10.16037/j.1007-869x.20255273
-
中图分类号/CLCN:U216.3
-
栏目/Col:土建工程
摘要:
［目的］在轨道交通检测监测中，轨道的精确里程信息对于检测数据的高效利用具有重要作用,因此需要研究并设计一种轨枕精确定位技术。［方法］由于轨道视频巡检系统采集图像并进行里程定位存在一定误差，故采用YOLOv8n目标检测模型对巡检图像的轨枕和地面电子标签进行定位和里程校正。基于YOLOv8n模型，使用全新的（增强交并比）方法，通过损失函数改进和轨枕计数的结构约束优化提出了一种基于改进YOLOv8n模型的轨枕目标检测模型YOLOv8n_SC。创新地引入结构优化算法，解决了轨枕被分割在前后2张图像内的计数问题，对重复检测和漏检问题进行结构化约束补齐。以广州某市域地铁线路为例，采用改进后的YOLOv8n_SC模型对线路轨枕和电子标签进行检测和里程校正，极大改善了对轨枕的漏检和误检。［结果及结论］经上述改进和约束后的轨枕精确定位模型YOLOv8n_SC显著提高了轨枕的定位精度，实现了轨枕级的精确里程定位方法，轨枕定位准确率达100%。在不增加系统设备情况下，YOLOv8n_SC模型提高了巡检图像采集系统的里程精度，使系统具有较好的可实施性。
Abstracts:
［Objective］ In rail transit inspection and monitoring, accurate mileage information of the track plays a critical role in the efficient utilization of inspection data. Therefore, it is necessary to study and design a precise sleeper localization technology. ［Method］ Due to certain errors in mileage positioning from images captured by track video inspection systems, the YOLOv8n object detection model is adopted to locate sleepers and ground electronic tags in inspection images and perform mileage correction. Based on the YOLOv8n model, a new EIoU (enhanced intersection over union) method is used. Through optimization of loss function and structural constraints on sleeper counting, a sleeper object detection model YOLOv8n_SC is proposed based on the improved YOLOv8n model. A structural optimization algorithm is innovatively introduced, and the sleeper counting issue when sleepers are split across two adjacent images is solved, and structural constraints are provided to mitigate missed and duplicate detections. Taking a city metro line in Guangzhou as an example, the improved YOLOv8n_SC model is applied to detect sleepers and electronic tags along the line and perform mileage correction. Thus, the scenarios of sleeper positioning missed and false detections are greatly improved. ［Result & Conclusion］ The improved and constrained sleeper positioning model YOLOv8n_SC significantly enhances sleeper positioning accuracy, achieving a sleeper-level precise mileage positioning method with 100% accuracy. Without requiring additional system equipment, the YOLOv8n_SC model improves the mileage accuracy of the inspection image acquisition system, providing strong practical feasibility.
过刊浏览
资料下载
友情链接