Point Cloud-based 3D Underwater Pose Estimation Using RANSAC and VFH Descriptors

Underwater pose estimation plays an important role in the process of underwater positioning and operation. In this paper, the point cloud data are collected by a depth camera, and the obtained point cloud data are clustered by RanSanc algorithm to accurately identify the 3D point cloud data of the target. By extracting the view feature histogram(VFH) of the target 3D point cloud data for subsequent pose estimation research, the time-consuming and labor-consuming caused by the large amount of overall point cloud data is avoided. Then, the VFH descriptors in different pose are trained and calibrated by the two-dimensional code truth measurement system, and the training set is saved by using the kd-tree neighbor search structure. Finally, the accuracy and feasibility of the proposed pose estimation algorithm are verified in a water tank experiments.