ElasticFusion: Dense SLAM Without A Pose Graph

By Thomas Whelan, Stefan Leutenegger, Renato F. Salas-Moreno, Ben Glocker, Andrew J. Davison. 2015

906 citations recorded by the source corpus.

We present a novel approach to real-time dense visual SLAM. Our system is capable of capturing comprehensive dense globally consistent surfel-based maps of room scale environments explored using an RGB-D camera in an incremental online fashion, without pose graph optimisation or any postprocessing steps. This is accomplished by using dense frame-tomodel camera tracking and windowed surfel-based fusion coupled with frequent model refinement through non-rigid surface deformations. Our approach applies local model-to-model surface loop closure optimisations as often as possible to stay close to the mode of the map distribution, while utilising global loop closure to recover from arbitrary drift and maintain global consistency.

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

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

ScanNet: Richly-Annotated 3D Reconstructions of Indoor Scenes (2017, background): “There is a large variety of algorithms targeting this scenario [59, 88, 7, 62, 37, 89, 42, 9, 90, 38, 12].”
OctNetFusion: Learning Depth Fusion from Data (2017, background): “This has led to the creation of large databases of real-world 3D content [7, 10, 42, 43], enabling progress in a variety of areas including 3D modeling [28, 54], 3D reconstruction [6, 52], 3D recognition [18, 44] and 3D scene understanding [16, 42].”
Shape Completion Using 3D-Encoder-Predictor CNNs and Shape Synthesis (2016, background): “Since the introduction of commodity range sensors such as the Microsoft Kinect, RGB-D scanning has gained a huge momentum in both ofﬂine and real-time contexts [26, 3, 28, 42, 4, 7].”
GS-SLAM: Dense Visual SLAM with 3D Gaussian Splatting (2023, methodology): “Traditional SLAM methods employ point/surfel clouds [18, 29, 39, 43], mesh representations [23], voxel hashing [10, 16, 21] or voxel grids [19] as scene representations to construct dense mapping, and have made considerable progress on localization accuracy.”
SemanticFusion: Dense 3D semantic mapping with convolutional neural networks (2016, background): “Our SemanticFusion pipeline is composed of three separate units; a real-time SLAM system ElasticFusion, a Convolutional Neural Network, and a Bayesian update scheme, as illustrated in Figure 2.”
Deep Global Registration (2020, result): “As shown in Table 2, compared to state-of-the-art online SLAM [42, 20, 38] and offline reconstruction methods [49], our approach yields consistently low error across scenes.”
Visual SLAM algorithms: a survey from 2010 to 2016 (2017, methodology): “In [66], deformation graph optimization is frequently used for certain frames, and camera motion is estimated by matching between a RGB-D image and a reconstructed model.”
Kimera: an Open-Source Library for Real-Time Metric-Semantic Localization and Mapping (2019, background): “SemanticFusion [17] RGB-D [18] surfels 3”
MaskFusion: Real-Time Recognition, Tracking and Reconstruction of Multiple Moving Objects (2018, methodology): “[50] presented a surfel-based RGBD-SLAM system for large environments with local and global loop-closure.”
Atlas: End-to-End 3D Scene Reconstruction from Posed Images (2020, background): “However, surfels (oriented point clouds) are starting to gain popularity [38,48].”
Scan2CAD: Learning CAD Model Alignment in RGB-D Scans (2018, background): “An alternative fusion approach is based on points [22]; the reconstruction quality is slightly lower, but it has more flexibility when handling scene dynamics and can be adapted on-the-fly for loop closures [40].”
EM-Fusion: Dynamic Object-Level SLAM With Probabilistic Data Association (2019, methodology): “Extensive research has gone into scaling the approaches to large environments [25, 14] or supporting loop-closing [10, 26] to reduce drift.”
Self-Supervised Visual Descriptor Learning for Dense Correspondence (2017, background): “A notable exception is ElasticFusion [28], which uses random ferns.”
PointDSC: Robust Point Cloud Registration using Deep Spatial Consistency (2021, neutral): “) Living1 Living2 Office1 Office2 AVG ElasticFusion [64] 66.”
Deep Local Shapes: Learning Local SDF Priors for Detailed 3D Reconstruction (2020, methodology): “of implicit SDFs in regular grids. This technique was further extended into real-time systems [30,45,35,36]. Surfaces are also shown to be represented by surfels, i.e. oriented planar surface patches [39,29,50]. Global approaches. Global implicit function approximation methods aim to approximate sin”
Co-SLAM: Joint Coordinate and Sparse Parametric Encodings for Neural Real-Time SLAM (2023, background): “Subsequent works focused on addressing the scalability issue by adopting more efficient data structures, such as surface elements (surfels) [35, 36], VoxelHashing [3, 9, 19] or Octrees [31,41].”
Fusion++: Volumetric Object-Level SLAM (2018, methodology): “not achieve state-of-the-art performance on these sequences such as [38, 22], and would require additional work, such as including joint depth and photometric tracking, to become competitive.”