首页|跨模态特征融合模板驱动的盾构隧道管片孔缝智能识别

跨模态特征融合模板驱动的盾构隧道管片孔缝智能识别

扫码查看
原文链接
  • 万方数据
  • 维普
基于点云的隧道数据处理方法已逐步替代人工用于盾构管片接缝识别中.但常规点云处理算法提取螺栓孔和接缝时会存在因附属设施遮挡造成错误识别的问题,导致目标识别精度降低.针对此问题,以广州2003年开通的某地铁线路某段地铁隧道数据为例,提出一种跨模态特征融合模板驱动的盾构隧道孔缝识别方法.首先将隧道断面点云的几何中心作为视点,以扫描测线为单元进行逐测线投影生成隧道二维图像;然后通过Canny边缘检测和Hough变换缓冲识别纵缝,对二维图像进行环片分割,并基于双模板匹配实现盾构环分类,依据环片模板进行螺栓孔和横缝的粗定位;最后利用螺栓孔点云DBSCAN聚类后的中心坐标对环片模板进行精校正,实现盾构隧道管片孔缝智能识别.研究结果表明:该方法在附属设施遮挡干扰等情况下能较好地实现孔缝精确识别.其中,设计的基于局部形态特征双模版驱动的盾构环片分类方法,可实现盾构环片的精确分类;设计的顾及盾构管片空间位置关系的孔缝识别方法,可有效提升孔缝识别精度.在识别率与耗时相近的情况下,本方法比同类方法识别精度更高,平均偏差更少,具有更好的准确性和鲁棒性.该方法将盾构隧道三维点云和二维点云投影图像进行数据特征融合的同时,能够顾及局部形态特征凭借双模板实现盾构环片的精确分类以及依据盾构管片的空间位置关系进一步提高孔缝识别精度.研究结果为进一步自动化精准识别盾构隧道接缝和螺栓孔目标信息提供参考.
Intelligent recognition of shield tunnel pipe sheet bolt holes and seam driven by cross-modal features and templates
The manual method of shield segment joint identification has been gradually superseded with a point cloud-based data processing technology.Nevertheless,there will be issues with incorrect or incapable identification brought on by the obstruction of auxiliary facilities when the traditional point cloud processing technique extracts bolt holes and seams,which will lower the accuracy of target recognition.To address this issue,a cross-modal feature fusion template-driven hole and joint identification approach for shield tunnels was presented,using as an example the data of a metro tunnel segment that opened in Guangzhou in 2003.Firstly,using the scanning survey line as the unit,a two-dimensional image of the tunnel was created by projecting the geometric center of the point cloud of the tunnel section as the viewpoint.The two-dimensional image was then segmented,the longitudinal seam was found using Canny edge detection and a Hough transform buffer,the shield ring classification was achieved using double template matching,and the transverse seam and bolt hole coarse positioning was done in accordance with the ring template.In order to achieve the intelligent identification of segment holes and seams in shield tunnels,the ring template was finally precisely corrected using the central coordinates of the bolt hole point cloud DBSCAN clustering.The results are drawn as follows.The proposed method can accurately identify holes and seams in the case of ancillary facilities occlusion interference.Accurate classification of shield rings can be achieved by the dual template-driven designed classification method that considers local morphological features.By taking into account the spatial position relationship of shield segments,the suggested hole and crack identification technique may effectively increase the accuracy of hole and crack identification.The proposed approach outperforms comparable methods in terms of accuracy and resilience,with less average variation,higher recognition accuracy,and better recognition rates and times.By combining the data features from the 3D point cloud and the 2D point cloud projection images of the shield tunnel,the method can accurately classify the shield ring by using a double template,account for local morphological features,and enhance the accuracy of hole and crack recognition based on the spatial position relationship of the shield segment.The results can serve as a guide for future automated and precise target information identification of bolt holes and shield tunnel seams.

identification of shield tunneling bolt hole and seamfeature fusiontemplate-drivencross-modalmobile laser scanning point cloud

谭兆、高贤君、杨元维、王少宁

展开 >

中国铁路设计集团有限公司,天津 300251

长江大学 地球科学学院,湖北 武汉 430100

盾构隧道孔缝识别 特征融合 模板驱动 跨模态 移动激光扫描点云

城市轨道交通数字化建设与测评技术国家工程实验室开放课题基金资助项目城市轨道交通数字化建设与测评技术国家工程实验室开放课题基金资助项目自然资源部环鄱阳湖区域矿山环境监测与治理重点实验室开放基金资助项目湖南科技大学测绘遥感信息工程湖南省重点实验室开放基金资助项目

2023ZH012021ZH02MEMI-2021-2022-08E22205

2024

10.19713/j.cnki.43-1423/u.T20231932

铁道科学与工程学报
中南大学 中国铁道学会

铁道科学与工程学报

CSTPCD北大核心EI
影响因子：0.837
ISSN：1672-7029
年,卷(期)：2024.21(9)
  • 15