Bridge substructure feature extraction based on the underwater sonar point cloud data

Obtaining the complete morphology of a bridge's substructure is an important prerequisite for evaluating the safety of wading bridges. However, due to the complex underwater environment, numerous noises, and lack of data, the detection of the substructure is still a major challenge. Work related to underwater scenes. To solve this problem, a solution based on the minimum cut optimization method is proposed for underwater sonar point cloud data (USPCD) to separate the pile group precisely. Considering the sparsity of USPCD, this method utilizes the spatial similarity of pile groups and first defines piles as source piles and target piles. The single pile segmentation parameters are first investigated on the source pile and then transferred to the target piles for coarse extraction, including the prior point and optimal sharing parameters. Then, based on the segmented piles, a 2D circular boundary fitting method is further developed for fine edge extraction to obtain piles with precisely defined boundaries. The performance of our proposed method is evaluated using the USPCD of the Wuhu Yangtze River Bridge. Results show that the pile foundation can be separated precisely, with recall and F1 -scored values of 0.974 and 0.835, respectively. Compared to other segmentation methods, this method provides good segmentation results. Furthermore, the influence of USPCD sparsity is also investigated, and results show that our proposed method can extract piles robustly even when the density is reduced by 28.3%. Therefore, the proposed pile point data extraction method can provide data support for the subsequent novelty detection and quantification of the pile foundation.