摘要
铁路隧道是铁路线路的关键组成部分,维持隧道环境的安全和稳定对保障列车正常运行具有重大意义。在铁路隧道的安全监测中,图像数据的分析和处理占据重要地位,研究提出一种基于深度神经网络的铁路隧道全景图像拼接算法,将多张具有重叠区域的图像拼接成大视场全景图。该算法通过将原始图像与蒙版相结合构建无先验信息子图数据集,训练深度神经网络学习最优参数,获得最优单应矩阵计算方法,并利用单应矩阵进行图像配准。此外,通过Voronoi规划寻找最佳拼接缝,并利用曝光补偿算法消除拼接图像亮度差异。在实际隧道数据集上,算法的性能指标RMSE、PSNR和SSIM分别达到2.94、26.59和0.806,均优于其他图像拼接算法,表明算法的有效性。
Railway tunnels are a key component of railway lines,and maintaining the safety and stability of the tunnel environment is of great significance for ensuring the normal operation of trains.In the safety monitoring of railway tunnels,the analysis and processing of image data play a crucial role.This study proposes a panoramic image stitching algorithm for railway tunnels based on deep neural networks,which stitches multiple images with overlapping areas into a wide-field panoramic image.This algorithm constructs a subgraph dataset without prior information by combining the original image with a mask,trains a deep neural network to learn optimal parameters,obtains the optimal homography matrix computation method,and utilizes the homography matrix for image registration.In addition,the Voronoi-based planning is used to find the optimal stitching seams,and an exposure compensation algorithm is applied to eliminate brightness differences in the stitched images.On the real tunnel dataset,the performance indicators RMSE,PSNR,and SSIM of the algorithm have reached 2.94,26.59,and 0.806,respectively,all of which are better than other image stitching algorithms,indicating the effectiveness of the algorithm.
作者
陈晓冈
CHEN Xiaogang(Ningbo Track Maintenance Depot,China Railway Shanghai Group Co.,Ltd.,Ningbo Zhejiang 315000,China)
出处
《中国铁路》
北大核心
2025年第8期53-58,共6页
China Railway
基金
中国铁路上海局集团有限公司科研计划项目(2021023)。
关键词
铁路隧道
图像拼接
深度神经网络
全景图像
拼接算法
railway tunnel
image stitching
deep neural network
panoramic image
stitching algorithm
