Abstract: High-precision 3D mapping is an important support for indoor 3D mapping. 3D mapping based on 3D LiDAR scanning technology is costly, requires the advance arrangement of targets, is prone to incomplete data in complex indoor environments, takes a long time to model 3D reconstruction based on image sequences, and is susceptible to a variety of factors. To address these problems, this paper applies RGB-D SLAM technology to indoor high-precision 3D mapping. By combining the depth camera with SLAM technology, the camera poses are calculated and 3D spatial information is recovered to obtain indoor 3D point cloud models, and the dense point cloud accuracy is evaluated with the actual measurement of the target object as the benchmark. The experimental results show that the method can quickly obtain high-precision 3D point cloud models with low cost and high efficiency, which can well meet the application requirements.

Key words: deep cameras, simultaneous localization and mapping(SLAM), 3D mapping, loopback detection, back-end optimisation