一种基于光度信息和ORB特征的建图SLAM

SLAM with Mapping Based on Photometric Information and ORB Features

针对同步定位与建图(Simultaneous Localization and Mapping,SLAM)系统经常因为相机抖动及场景结构单一等问题而导致跟踪失败的情况,以及重建三维地图的任务,提出了一种混合SLAM方法(R-ORB SLAM)。该方法采用一种基于光度误差的位姿粗略估计作为特征视觉里程计的先验,在ORB-SLAM2跟踪失败的情况下,使用该结果参与位姿估计。同时,在建图模块,对由每个关键帧点云拼接得到的全局点云地图使用VoxelGrid滤波器进行下采样,得到去除冗余点的稠密三维点云地图。通过对三维点云地图使用Poisson算法实现表面重建,得到三维地图模型。在两个流行的公开数据集上的实验结果表明,本文方法能有效解决跟踪失败问题,实现三维地图重建,具有较高的跟踪精度与重建精度。

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.