联合MSF和FCD的公路隧道视频裂缝关键帧提取算法  

Video keyframe extraction algorithm for highway tunnel cracks based on the combination of MSF and FCD

作　　者：王萍[1,2] 秦川 朱军 刘洋[3,4,5] 谢亚坤[1] 孙中秋 赖建波 党沛 WANG Ping;QIN Chuan;ZHU Jun;LIU Yang;XIE Yakun;SUN Zhongqiu;LAI Jianbo;DANG Pei(Faculty of Geosciences and Engineering,Southwest Jiaotong University,Chengdu 610000,China;Sichuan Transportation Survey and Design Institute Co.,Ltd.,Chengdu 610000,China;Guangzhou Urban Planning&Design Survey Research Institute,Guangzhou 510060,China;Guangzhou Resource Planning and Marine Science and Technology Collaborative Innovation Center,Guangzhou 510060,China;Guangdong Enterprise Key Labora-tory for Urban Sensing,Monitoring and Early Warning,Guangzhou 510060,China)

机构地区：[1]西南交通大学地球科学与工程学院,成都610000 [2]四川省交通勘察设计院有限公司,成都610000 [3]广州市城市规划勘测设计研究院,广州510060 [4]广州市资源规划和海洋科技协同创新中心,广州510060 [5]广东省城市感知与监测预警企业重点实验室,广州510060

出　　处：《北京交通大学学报》2024年第5期98-106,共9页JOURNAL OF BEIJING JIAOTONG UNIVERSITY

基　　金：国家自然科学基金(42271424,42171397);四川省交通勘察设计研究院有限公司自立科技项目(232023004);国家留学基金资助项目(202307000096)。

摘　　要：针对公路隧道视频关键帧提取精度低和速度慢的问题,提出一种联合多尺度滤波(Multi-Scale Filtering,MSF)和脊变化检测(Ridge Change Detection,RCD)的公路隧道视频裂缝关键帧高效提取算法.首先,基于多尺度滤波和Hessian矩阵设计裂缝脊特征提取方法,考虑裂缝在不同方向和尺度下的梯度和二阶导数性质,通过特征值解算和阈值分析,提取和融合不同尺度滤波结果中的脊线,实现公路隧道视频裂缝脊特征准确提取;然后,提出一种道路裂缝视频帧的索引空间模型,基于脊线差分分析和帧间相异约束,构建裂缝关键帧索引机制,利用脊线变化检测裂缝区域的动态特征,并通过帧间相异度判别筛选出具有代表性的关键帧,从而减少冗余帧,显著提高裂缝检测视频的处理效率;最后,开展公路隧道裂缝视频关键帧提取实验.实验结果表明:所提方法平均准确率较现有关键帧提取方法提高19.3%~43.2%,裂缝关键帧平均提取速度是基于运动的方法的11~13倍,有效提高了公路隧道裂缝检测效率,能够为公路隧道裂缝智能检测提供参考.To address the issues of low accuracy and slow speed in extracting keyframes from highway tunnel videos,this study proposes an efficient extraction algorithm for crack keyframe extraction,integrating Multi-Scale Filtering(MSF)and Ridge Change Detection(RCD).First,a crack ridge feature extraction method is developed based on multi-scale filtering and the Hessian matrix,considering the gradient and second-order derivative properties of cracks across various directions and scales.Using eigenvalue computation and threshold analysis,ridge lines from different scale filtering results are extracted and fused,enabling precise extraction of crack ridge features in highway tunnel videos.Then,an indexing spatial model for road crack video frames is proposed.By employing ridge differential analysis and inter-frame dissimilarity constraints,a crack keyframe indexing mechanism is constructed.Dynamic characteristics of crack regions are identified through ridge change detection,while representative keyframes are selected using inter-frame dissimilarity discrimination.This approach reduces redundant frames and significantly improves the efficiency of video processing for crack detection.Finally,experiments are conducted to extract keyframes from highway tunnel crack videos.The experimental results demonstrate that the proposed method improves accuracy by 19.3%to 43.2%compared to existing keyframe extraction methods,with average keyframe extraction speeds being 11 to 13 times faster than motion-based methods.This method effectively enhances highway tunnel crack detection efficiency and provides valuable insights for intelligent crack detection in highway tunnels.

关 键 词：公路隧道视频 裂缝关键帧提取 多尺度滤波 脊变化检测 帧间相异度 

分 类 号：P232[天文地球—摄影测量与遥感]