FastVL: Fast Moving 3D Indoor Reconstruction System Based on SE(2) Constraint for Visual-Laser Fusion

Abstract

Three-dimensional reconstruction methods face challenges in computational power and robustness when dealing with fast-moving scenes. This paper proposes a fast moving three-dimensional (3D) indoor reconstruction system based on SE(2) constraint for visual-laser fusion. The system utilizes the fusion of visual and 2D laser data to acquire both color texture and more accurate camera pose information. To reduce the complexity of multi-dimensional motion computation, this paper introduces ground constraint to reduce the dimensionality from SE(3) to SE(2). The moving Truncated Signed Distance Function (TSDF) method effectively eliminates interference from discrete points in the model. By deploying a RGB-D camera and a 2D LiDAR on a differential drive platform, the high-precision 3D reconstruction performance of the proposed algorithm in fast-moving scenes is validated through comparisons with real-world datasets.