被动式三维重建在建立桥梁缆索数字模型中的应用

Application of Passive 3D Reconstruction in Establishing Digital Model of Bridge Cable

摘要: 在斜拉桥、拱式桥以及悬索桥等大跨径桥梁的建设中，缆索有着广泛的应用。缆索的有效承载能力会影响到桥梁的稳定和寿命，因此桥梁缆索的检测工作是一项十分重要的工作。建立缆索的数字模型可以方便后续的损伤检测工作，三维重建技术也因此在缆索检测问题上有着诸多应用。被动式三维重建是从一张或多张图片中恢复场景的几何结构，最终得到该场景的三维模型的方法。而通过被动式三维重建的方法，可以仅从桥梁缆索的影像数据来重建缆索的三维数字模型。本文使用缆索爬索检测机器人在移动过程中拍摄的影像数据进行实验，对获取的影像数据进行畸变矫正等数据处理；并根据桥梁缆索的几何结构分析得到相机位置信息，结合不同视点上的多幅图像恢复三维信息，最终得到三维模型。对最终重建模型选定4个点用作几何精度评定并与真值对比，选定点的相对误差分别达到了0.913%、1.196%、0.947%、0.568%，并得到平均绝对误差MAE为1.0053 mm。试验结果表明，本文提出方法可以在减少数据获取成本的同时得到纹理清晰、精度较高的模型。

Abstract: Cables are widely used in the construction of long-span bridges such as cable-stayed bridges, arch bridges and suspension bridges. The detection of the cables is a critical task, since the effective bearing area of the bridge cables will affect the stability and service life of the bridge. Establishing the digital model of the cable can facilitate the subsequent damage detection work. This led to many applications with 3D reconstruction technology in cable detection. Passive 3D reconstruction is a method to restore the geometric structure of a scene from one or more pictures, and finally obtain the 3D model of the scene. With the passive 3D reconstruction technology, the 3D digital model of the cable can be reconstructed only using the image data of bridge cables. In this work, the image data taken by the cable detection robot while moving is used for experiment, and data processing such as distortion correction is performed on the acquired image data. The information of camera position can be obtained based on the geometric structure of bridge cable. Combining this infor-mation with multiple images from different viewpoints, the three-dimensional information can be recovered for the final 3D model. For the final reconstructed model, 4 points were selected for geo-metric accuracy evaluation and compared with the ground truth. The relative errors of these chosen points reach 0.913%, 1.196%, 0.947% and 0.568% respectively, and the MAE is 1.0053 mm. The experimental results show that the method proposed can generate a model with clear texture and high accuracy, and also reduce the cost of data acquisition.