A high-quality voxel 3D reconstruction system for large scenes based on the branch and bound method

Abstract

The accuracy of three-dimensional (3D) reconstruction models is directly related to camera pose estimation. To obtain a high-quality 3D reconstruction model in a complex large-scale scene, this paper proposes a high-quality voxel 3D reconstruction system for large scenes based on the branch and bound method. Aiming at the problem of falling into local optimum in the 3D reconstruction system, a branch and bound method is proposed to avoid the influence of the local optimum on the model, and the optimal camera posse and corresponding relationship are given. After optimization of the high-precision camera pose, the voxels are used to reconstruct 3D models by using the truncated signed distance function. Finally, the high-precision 3D model is obtained. Many experimental data show that this method could reliably find the best pose data and has high precision dense 3D model. The proposed method can solve a variety of challenging large-scale scenes, such as fast-moving, long-loop closing, multiple rotation, and large-scale scenes. This method can be widely used in the fields of intelligent robots, robot navigation, unmanned driving, virtual reality, object recognition, and so on.