Real-time Underwater 3D Reconstruction Method Based on Stereo Camera

Abstract

Accurate 3D reconstruction of underwater scenes is very important in various fields related to underwater operation. Multi-beam bathymetry sonar can be used to reconstruct terrain map of seafloor. However, this 3D map is not able to express details of objects well. Accurate underwater targets can only be reconstructed through the camera. However, the existing research has problems such as sparse point cloud in real-time reconstruction and time-consuming for dense point cloud reconstruction. In this paper, we propose an underwater real-time 3D reconstruction system based on stereo camera. Aiming at the problem of blurred underwater images, we propose a lightweight image enhancement algorithm. For the problem of real-time point cloud generation, we adopt the method of computing disparity maps by using stereo cameras. We fuse edge information in the process of computing the disparity map, which improves the accuracy of the disparity map. For multi-view point cloud stitching, we propose a fusion stitching method based on SLAM system and dense point cloud. Our method has been tested in indoor pools and sea areas near the marina respectively. The experimental results show that our method has certain advantages in the 3D reconstruction effect and accuracy of underwater objects.