SLAM with Mapping Based on Photometric Information and ORB Features

Aiming at the tracking failure caused by camera jitter or low-texture environment in slam, this paper proposes a hybrid slam method R-ORB slam for depth camera to achieve the task of 3D map reconstruction. A rough pose estimation method based on photometric error is used as the prior of the feature-based odometer. Under the case that ORB-SLAM2 tracking fails, the result of the method is used to participate in the pose estimation. Meanwhile, for generating dense three-dimensional point map with non-redundant points, the VoxelGrid filter is used to down sample the global point cloud map obtained from each key frame point clouds mosaic. Then, by using Poisson algorithm to reconstruct the surface of 3D point cloud map, we can obtain the 3D map model. It is shown via the experiments on two popular open datasets that the proposed method can effectively solve the problem of tracking failure and realize 3D reconstruction with high tracking accuracy and reconstruction accuracy.