Point Cloud Method for Detecting Suspended Pipelines Using Multi-Beam Water Column Data

In the task of inspecting underwater suspended pipelines, multi-beam sonar (MBS) can provide two-dimensional water column images (WCIs). However, systematic interferences (e.g., sidelobe effects) may induce misdetection in WCIs. To address this issue and improve the accuracy of detection, we developed a density-based clustering method for three-dimensional water column point clouds. During the processing of WCIs, sidelobe effects are mitigated using a bilateral filter and brightness transformation. The cross-sectional point cloud of the pipeline is then extracted by using the Canny operator. In the detection phase, the target is identified by using density-based spatial clustering of applications with noise (DBSCAN). However, the selection of appropriate DBSCAN parameters is obscured by the uneven distribution of the water column point cloud. To overcome this, we propose an improved DBSCAN based on a parameter interval estimation method (PIE-DBSCAN). First, kernel density estimation (KDE) is used to determine the candidate interval of parameters, after which the exact cluster number is determined via density peak clustering (DPC). Finally, the optimal parameters are selected by comparing the mean silhouette coefficients. To validate the performance of PIE-DBSCAN, we collected water column point clouds from an anechoic tank and the South China Sea. PIE-DBSCAN successfully detected both the target points of the suspended pipeline and non-target points on the seafloor surface. Compared to the K-Means and Mean-Shift algorithms, PIE-DBSCAN demonstrates superior clustering performance and shows feasibility in practical applications.