Zou Yajing, Eldemiry Amr, Li Yaxin, Chen Wu
Shenzhen Research Institute, The Hong Kong Polytechnic University, Shenzhen 518057, China.
Department of Land Surveying and Geo-Informatics, The Hong Kong Polytechnic University, Hong Kong 999077, China.
Sensors (Basel). 2020 Sep 2;20(17):4984. doi: 10.3390/s20174984.
Three-dimensional (3D) reconstruction using RGB-D camera with simultaneous color image and depth information is attractive as it can significantly reduce the cost of equipment and time for data collection. Point feature is commonly used for aligning two RGB-D frames. Due to lacking reliable point features, RGB-D simultaneous localization and mapping (SLAM) is easy to fail in low textured scenes. To overcome the problem, this paper proposes a robust RGB-D SLAM system fusing both points and lines, because lines can provide robust geometry constraints when points are insufficient. To comprehensively fuse line constraints, we combine 2D and 3D line reprojection error with point reprojection error in a novel cost function. To solve the cost function and filter out wrong feature matches, we build a robust pose solver using the Gauss-Newton method and Chi-Square test. To correct the drift of camera poses, we maintain a sliding-window framework to update the keyframe poses and related features. We evaluate the proposed system on both public datasets and real-world experiments. It is demonstrated that it is comparable to or better than state-of-the-art methods in consideration with both accuracy and robustness.
使用具有同步彩色图像和深度信息的RGB-D相机进行三维（3D）重建很有吸引力，因为它可以显著降低设备成本和数据采集时间。点特征通常用于对齐两个RGB-D帧。由于缺乏可靠的点特征，RGB-D同步定位与地图构建（SLAM）在低纹理场景中很容易失败。为了克服这个问题，本文提出了一种融合点和线的鲁棒RGB-D SLAM系统，因为当点不足时，线可以提供鲁棒的几何约束。为了全面融合线约束，我们在一个新颖的代价函数中结合了二维和三维线重投影误差与点重投影误差。为了解代价函数并滤除错误的特征匹配，我们使用高斯-牛顿法和卡方检验构建了一个鲁棒的位姿求解器。为了校正相机位姿的漂移，我们维护一个滑动窗口框架来更新关键帧位姿和相关特征。我们在公共数据集和实际实验上对所提出的系统进行了评估。结果表明，在准确性和鲁棒性方面，它与现有最先进的方法相当或更好。
