基于三维图像的铁路扣件缺陷自动识别算法  

Automatic Defect Inspection Algorithm of Railway Fasteners Based on 3D Images

作　　者：代先星 阳恩慧[1,2] 丁世海[1,2] 王郴平[1,2,3] 邱延峻 

机构地区：[1]西南交通大学土木工程学院,四川成都610031 [2]西南交通大学道路工程四川省重点实验室,四川成都610031 [3]俄克拉荷马州立大学土木与环境工程学院,俄克拉荷马ok74078 [4]西南交通大学高速铁路线路工程教育部重点实验室,四川成都610031

出　　处：《铁道学报》2017年第10期89-96,共8页Journal of the China Railway Society

基　　金：国家自然科学基金(51478398;51308477;U1534203);中央高校基本科研业务费专项资金(2682015CX091)

摘　　要：针对当前铁路扣件状态自动识别准确率和稳定性不高等问题,利用直射式激光三角测量法原理研发扣件检测系统,采集不受环境光影响的高质量轨道三维数据。提出基于三维图像的扣件区域定位方法,并利用先验知识验证扣件位置以保证扣件定位的准确性;基于弹条的高度规律信息提取弹条,采用HGOH作为特征描述算子;根据特征向量的模是否等于零可识别出缺失扣件,将模不为零的特征向量送入已训练的SVM分类器,从而识别断裂扣件和完整扣件。室内试验研究结果表明,采用本文提出的扣件缺陷自动检测算法,识别准确率可达98.0%,能满足扣件缺陷自动化检测的需要。Aiming at the low recognition accuracy and stability of detection algorithms for railway fastener de-fects ,a novel fastener inspection system based on direct-type laser triangulation was developed to collect high quality three-dimensional railway data without the effects of environmental illumination. The effective fastener location algorithm based on 3D images verifying the location of fastener with priori-knowledge was presented to achieve high locating accuracy. The clip sub-images were extracted correctly from fastener region images based on the height characteristics of the clip. The method of Height Gradients Oriented Histogram（ HGOH） was applied to extract the feature of images. If the norm of feature vector was zero, the image would be recognized as the missing fastener. Otherwise the feature vectors whose norm of the vector was not zero were fed into the trained Support Vector Machine（SVM） classifier to distinguish the broken and intact fasteners. The experi-mental results indicate that the average recognition rate of the proposed automatic inspection algorithm of fastener defects is 98. 0 % , w h i c h can m e e t the requirement of automatic fastener detection.

关 键 词：三维激光 铁路扣件检测 识别算法 准确率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]