期刊文献+

钢桥密集螺栓异常状态视觉识别方法 被引量:2

Visual Recognition Method for Abnormal States of Dense Bolts for Steel Bridges
下载PDF
导出
摘要 针对传统机器视觉方法无法从不同拍摄视角和拍摄距离的图像中较好地识别出异常螺栓的问题,依据钢桥密集螺栓区域自身视觉特点,提出基于区域异常点分析的密集螺栓异常状态视觉识别方法。该方法首先提取和比对图像蓝色和红色通道灰度,完成梁体颜色分割,并运用选定的Canny算子提取梁体区域内边缘,采用Hough线识别方法剔除杂波;其次依据密集螺栓呈现簇状分布特点,运用密度聚类分析定位螺栓簇区域,并依据密集螺栓位置呈现平行网格分布的特点,运用投影分析定位单个螺栓区域;然后依据各螺栓的阴影特征,利用切比雪夫不等式快速判定螺栓状态,完成螺栓异常识别;最后,制作钢桥节点板模型,采集不同螺栓松动或脱落图像,对该方法进行测试。结果表明:该方法对图像拍摄视角和距离的适用度高,对螺栓脱落的识别能力优于对螺栓松动的识别;不同场景下单个螺栓定位的平均交并比大于0.75,且螺栓脱落和松动识别的准确率和召回率分别在0.89和0.85以上。 In view of the issue that traditional machine vision methods cannot identify abnormal bolts well in the images taken from different shooting angles and distances,a vision recognition method for abnormal states of dense bolts based on regional abnormal point analysis is proposed according to the visual characteristics of the dense bolt areas in steel bridges.Firstly,this method extracts and compares the grayscale of the blue and red channels in the image to complete the color segmentation of the beam body.It then selects the Canny operator to extract the inner edge of the beam body area and uses the Hough line recognition method to eliminate clutter.Secondly,based on the cluster distribution characteristics of dense bolts,the bolt cluster area is located by clustering density analysis,and based on the parallel grid distribution characteristics of dense bolt positions,the single bolt area is located by projection analysis.Finally,according to the shadow features of each bolt,Chebyshev′s inequality is used to quickly determine the bolt states and complete the recognition of abnormal bolts.A gusset plate model of the steel bridge is created,and different images of bolt loosening or falling off are collected for method testing.The results show that the method is highly applicable to image shooting angles and distances,and the recognition ability of bolt falling off is greater than that of bolt loosening.In different scenarios,the average intersection ratio of a single bolt location is greater than 0.75.The accuracy and recall rates of bolt falling off and loosening recognition are above 0.89 and 0.85,respectively.
作者 王保宪 欧丙泽 赵维刚 谭兆 秦守鹏 WANG Baoxian;OU Bingze;ZHAO Weigang;TAN Zhao;QIN Shoupeng(Structural Health Monitoring and Control Key Laboratory of Hebei Province,Shijiazhuang Tiedao University,Shijiazhuang Hebei 050043,China;State Key Laboratory of Mechanical Behavior and System Safety of Traffic Engineering Structures,Shijiazhuang Tiedao University,Shijiazhuang Hebei 050043,China;School of Electrical and Electronic Engineering,Shijiazhuang Tiedao University,Shijiazhuang Hebei 050043,China;National Engineering Research Center for Digital Construction and Evaluation Technology of Urban Rail Transit,Tianjin 300308,China)
出处 《中国铁道科学》 EI CAS CSCD 北大核心 2023年第5期81-93,共13页 China Railway Science
基金 国家重点研发计划课题(2022YFB2603303) 国家自然科学基金资助项目(52178293,51808358) 中国国家铁路集团有限公司实验室基础研究项目(L2021G013) 河北省自然科学基金创新研究群体项目(E2021210099) 河北省高等学校科技研究项目(BJ2020057)。
关键词 钢桥 螺栓 脱落 松动 视觉识别 密度聚类分析 投影分析 Steel bridges Bolts Falling off Loosening Visual recognition Clustering density analysis Projection analysis
  • 相关文献

参考文献9

二级参考文献67

共引文献126

同被引文献12

引证文献2

二级引证文献1

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部