Gabara Grzegorz, Sawicki Piotr
Institute of Geodesy, University of Warmia and Mazury in Olsztyn, 10-719 Olsztyn, Poland.
Sensors (Basel). 2018 Mar 6;18(3):791. doi: 10.3390/s18030791.
The paper presents the results of testing a proposed image-based point clouds measuring method for geometric parameters determination of a railway track. The study was performed based on a configuration of digital images and reference control network. A DSLR (digital Single-Lens-Reflex) Nikon D5100 camera was used to acquire six digital images of the tested section of railway tracks. The dense point clouds and the 3D mesh model were generated with the use of two software systems, RealityCapture and PhotoScan, which have implemented different matching and 3D object reconstruction techniques: Multi-View Stereo and Semi-Global Matching, respectively. The study found that both applications could generate appropriate 3D models. Final meshes of 3D models were filtered with the MeshLab software. The CloudCompare application was used to determine the track gauge and cant for defined cross-sections, and the results obtained from point clouds by dense image matching techniques were compared with results of direct geodetic measurements. The obtained RMS difference in the horizontal (gauge) and vertical (cant) plane was RMS∆ < 0.45 mm. The achieved accuracy meets the accuracy condition of measurements and inspection of the rail tracks (error m < 1 mm), specified in the Polish branch railway instruction Id-14 (D-75) and the European technical norm EN 13848-4:2011.
本文介绍了一种用于确定铁路轨道几何参数的基于图像的点云测量方法的测试结果。该研究基于数字图像和参考控制网络的配置进行。使用尼康D5100数码单反相机获取了铁路轨道测试段的六张数字图像。利用RealityCapture和PhotoScan这两个软件系统生成了密集点云和3D网格模型，这两个软件分别采用了不同的匹配和3D物体重建技术：多视图立体匹配和半全局匹配。研究发现，这两个应用程序都能生成合适的3D模型。3D模型的最终网格用MeshLab软件进行了滤波。使用CloudCompare应用程序确定定义横截面的轨距和超高，并将通过密集图像匹配技术从点云获得的结果与直接大地测量结果进行比较。在水平（轨距）和垂直（超高）平面上获得的均方根差值为RMS∆<0.45mm。所达到的精度符合波兰铁路支线指令Id-14（D-75）和欧洲技术规范EN 13848-4:2011中规定的铁路轨道测量和检查的精度条件（误差m<1mm）。
