Multisensor Fusion for 3-D Reconstruction of Urban Pipeline Interior Using the Enhanced R³LIVE Algorithm

Accurately obtaining the 3-D geometry and texture of urban pipeline interiors is crucial for pipeline inspection, urban planning, and digital management. The interior environment of urban pipelines is characterized by insufficient illumination and a lack of textural features, which poses significant challenges for real-time 3-D reconstruction. In this study, we propose a novel multisensor fusion system based on the enhanced R³LIVE algorithm, which integrates LiDAR, cameras, and inertial measurement unit (IMU). Our approach addresses the performance degradation of LiDAR and vision systems within pipelines by employing a hybrid optical-flow method based on the convolutional neural networks (CNNs), which significantly enhances the accuracy and efficiency of tracking feature points on the inner wall of pipeline. Quantitative experimental results obtained in a real urban pipeline environment demonstrate that the accuracy and robustness of the 3-D reconstruction are substantially improved, achieving a relative rotation error (RRE) of 1.75° and a relative translation error (RTE) of 2.41% over a range of 150 m. Moreover, the algorithm significantly outperforms the FAST-LIVO, FAST-LIO2, and the original R³LIVE conventional simultaneous localization and mapping (SLAM) methods at longer ranges. These results underscore the practical value and superiority of our multisensor fusion method for the 3-D reconstruction of urban pipelines.