无砟轨道板裂缝三维激光检测系统研发与算法验证  

Development and Algorithm Verification of 3D Laser Detection System for Non-ballasted Track Slab Cracks

作　　者：战友 阳恩慧[1] 马啸天 安哲立 代先星 王郴平[1,3] ZHAN You;YANG Enhui;MA Xiaotian;AN Zheli;DAI Xianxing;WANG Chenping(School of Civil Engineering,Southwest Jiaotong University,Chengdu 610031,China;Railway Engineering Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;School of Civil and Environment Engineering,Oklahoma State University,Stillwater 74078,USA)

机构地区：[1]西南交通大学土木工程学院,四川成都610031 [2]中国铁道科学研究院集团有限公司铁道建筑研究所,北京100081 [3]俄克拉何马州立大学土木与环境工程学院,俄克拉何马州静水74078

出　　处：《铁道学报》2021年第7期114-120,共7页Journal of the China Railway Society

基　　金：国家自然科学基金(52008354,U1534203);中国博士后科学基金(2019M663557);中央高校基本科研业务费(2682020CX65)。

摘　　要：针对高速铁路无砟轨道板的表面损伤问题,现有的检测系统与方法仍难以达到高精度、高速度、高准确度及自动化要求。随着高速微型计算机和三维激光扫描技术的发展,应用超高精度3D自动化检测系统来提升轨道板伤损识别精度和检测效率的趋势越发明显。在已有数字化信息采集与自动分析系统基础上自主研发无砟轨道板裂缝三维激光检测系统,通过融合自适应阈值分割、PGM概率图模型与SVM机器学习方法建立轨道板裂缝集成学习识别算法,准确、快速、智能识别轨道板裂缝。结果表明,该系统能够获取高精度轨道板三维图像数据,所提出的轨道板裂缝集成学习识别算法能达到较高的准确率(均值97.3%)和召回率(均值92.6%),综合表现优于传统算法。Aiming at the surface damage of non-ballasted track slab of high-speed railway,the existing detection systems and methods can hardly achieve high precision,high speed,high accuracy and automation performance.The development of high-speed microcomputer and 3D laser scanning technology results in wider application of ultra-high precision 3D automatic detection system to improve the accuracy and efficiency of crack identification.This study developed a 3D laser crack detection system for non-ballasted track slabs based on the research of existing digital information collection and automatic analysis system.Ensemble-learning based crack detection algorithm was developed by integrating adaptive threshold segmentation,PGM probability map model and SVM machine learning algorithms,to achieve fast and intelligent identification of railway slab cracks.The results indicate that the developed system could acquire high-quality 3D surface data of non-ballasted track slab.This paper achieved the pixel level recognition of crack by ensemble learning with relatively high accuracy(97.3%)and recall rate(92.6%).The proposed algorithm in this study outperforms the traditional crack identification algorithms.

关 键 词：三维激光 无砟轨道板 裂缝 集成学习 

分 类 号：U213.2[交通运输工程—道路与铁道工程]