BAD SLAM: Bundle Adjusted Direct RGB-D SLAM

By Thomas Schöps, Torsten Sattler, Marc Pollefeys. Computer Vision and Pattern Recognition, 2019

453 citations recorded by the source corpus.

A key component of Simultaneous Localization and Mapping (SLAM) systems is the joint optimization of the estimated 3D map and camera trajectory. Bundle adjustment (BA) is the gold standard for this. Due to the large number of variables in dense RGB-D SLAM, previous work has focused on approximating BA. In contrast, in this paper we present a novel, fast direct BA formulation which we implement in a real-time dense RGB-D SLAM algorithm. In addition, we show that direct RGB-D SLAM systems are highly sensitive to rolling shutter, RGB and depth sensor synchronization, and calibration errors. In order to facilitate state-of-the-art research on direct RGB-D SLAM, we propose a novel, well-calibrated benchmark for this task that uses synchronized global shutter RGB and depth cameras. It includes a training set, a test set without public ground truth, and an online evaluation service. We observe that the ranking of methods changes on this dataset compared to existing ones, and our proposed algorithm outperforms all other evaluated SLAM methods. Our benchmark and our open source SLAM algorithm are available at: www.eth3d.net

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition, Geology

View Paper: https://www.semanticscholar.org/paper/8ef6983143e8999595def5b668f2f0140fde1185

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computer Vision and Pattern Recognition

In Geology

Citations in the corpus

D3VO: Deep Depth, Deep Pose and Deep Uncertainty for Monocular Visual Odometry (2020, methodology): “…and the direct VO share a similar photometric objective, we propose to use the learned weights to replace the weighting function of the photometric residual in traditional direct VO which is empirically set [61] or only accounts for the intrinsic uncertainty of the speciﬁc algorithm itself [16,37].”
MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision (2024, unknown): “We evaluate the accuracy of point map estimation on eight diverse datasets: NYUv2 [35], KITTI [47], ETH3D [45], iBims-1 [29], Sintel [8], Google Scanned Objects (GSO) [15], DDAD [22], and DIODE [48].”
iMAP: Implicit Mapping and Positioning in Real-Time (2021, methodology): “We compare with surfel-based BAD-SLAM [24], TSDF fusion Kintinuous [38], and sparse ORB-SLAM2 [16], state-of-the-art SLAM systems.”
SplaTAM: Splat, Track & Map 3D Gaussians for Dense RGB-D SLAM (2023, background): “Traditional approaches to dense SLAM have explored a variety of explicit representations, including 2.5D images [15, 34], (truncated) signed-distance functions [6, 25, 26, 42], gaussian mixture models [9, 10] and circular sur-fels [13, 33, 43].”
GS-SLAM: Dense Visual SLAM with 3D Gaussian Splatting (2023, methodology): “BAD-SLAM [26] proposes to jointly optimize the keyframe poses and 3D scene geometry via a direct bundle adjustment (BA) technique.”
Gaussian Splatting SLAM (2023, unknown): “Such flexibility benefits online applications such as SLAM with deformation-based loop closure [29, 41].”
NICE-SLAM: Neural Implicit Scalable Encoding for SLAM (2021, methodology): “On the other hand, the world-centric map representation anchors the 3D geometry in uniform world coordinates, and can be further divided into surfels [42, 58] and voxel grids, typically storing occupancies or TSDF values [11].”
Deep Global Registration (2020, background): “As shown in Table 2, compared to state-of-the-art online SLAM [43, 21, 39] and ofﬂine reconstruction methods [50], our approach yields consistently low error across scenes.”
Metric3D v2: A Versatile Monocular Geometric Foundation Model for Zero-Shot Metric Depth and Surface Normal Estimation (2024, unknown): “These unique attributes render both depth and surface normals indispensable in various computer vision applications, including 3D reconstruction [1], [2], [3], neural rendering (NeRF) [4], [5], [6], [7], autonomous driving [8], [9], [10], and robotics [11], [12], [13].”
DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras (2021, methodology): “ETH-3D RGB-D SLAM leaderboard [41], outperforming the second place by 35% under the AUC metric which considers both error and rate of catastrophic failure.”
H2O: Two Hands Manipulating Objects for First Person Interaction Recognition (2021, methodology): “As training data, we use the masks obtained by projecting our 3D models onto the images used for their BADSLAM-based reconstruction.”
Vox-Fusion: Dense Tracking and Mapping with Voxel-based Neural Implicit Representation (2022, methodology): “Following this basic design, many systems gain improvements by introducing different 3D structures [38], exploring space subdivision [21, 24] and performing global map optimization [11,13,26].”