Real-Time 3D Reconstruction in Dynamic Scenes Using Point-Based Fusion

By M. Keller, Damien Lefloch, M. Lambers, Shahram Izadi, Tim Weyrich, Andreas Kolb. International Conference on 3D Vision, 2013

409 citations recorded by the source corpus.

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

View Paper: https://www.semanticscholar.org/paper/2350da8e3da215c6fddf149a081442ebabaf8a44

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

BAD SLAM: Bundle Adjusted Direct RGB-D SLAM (2019, methodology): “Examples include voxel-based KinectFusion-style approaches [28, 45, 47], similar methods based on surfel representations [29, 35, 36], as well as some keyframe-based techniques [39–43].”
MaskFusion: Real-Time Recognition, Tracking and Reconstruction of Multiple Moving Objects (2018, methodology): “We adopt the surfel representation popularised by [23, 50], where a model Mm is represented by a cloud of surfels M s m ∈ (p ∈ R3,n ∈ R3,c ∈ N3,w ∈ R,r ∈ R, t ∈ R2) ∀s < |Mm|, which are tuples of position, normal, colour, weight, radius and two timestamps.”
Deeper Depth Prediction with Fully Convolutional Residual Networks (2016, methodology): “In particular, we deploy a SLAM framework where frame-to-frame tracking is obtained via GaussNewton optimization on the pixelwise intensity differences computed on consecutive frame pairs as proposed in [12], while fusion of depth measurements between the current frame and the global model is carrie”
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].”
ElasticFusion: Dense SLAM Without A Pose Graph (2015, methodology): “This is well within the widely accepted minimum frequencies for fused dense SLAM algorithms [24, 17, 2, 9], and as shown in our qualitative results more than adequate for real-time operation.”
iMAP: Implicit Mapping and Positioning in Real-Time (2021, background): “Some approaches in dense SLAM explicitly represent surfaces [8, 37], but direct representation of volume is desirable to enable a full range of applications such as planning.”
Kimera: an Open-Source Library for Real-Time Metric-Semantic Localization and Mapping (2019, neutral): “…surfels Voxblox [27] RGB-D [26] TSDF SLAM++ [16] RGB-D alternation objects SemanticFusion [17] RGB-D [18] surfels Mask-fusion [28] RGB-D [29] surfels SegMap [30] lidar GTSAM points/segments XIVO [31] mono/IMU EKF objects Voxblox++ [14] RGB-D [26] TSDF Kimera mono/stereo/IMU GTSAM…”
EM-Fusion: Dynamic Object-Level SLAM With Probabilistic Data Association (2019, background): “alignment [4], and map representations such as surfels [9] or keyframes [10].”
VolumeDeform: Real-Time Volumetric Non-rigid Reconstruction (2016, methodology): “Online Static Reconstruction : Methods for ofﬂine static 3D shape reconstruction from partial RGB-D scans differ in the employed scene representation, such as point-based representations [19–21] or meshes [22].”
ReFusion: 3D Reconstruction in Dynamic Environments for RGB-D Cameras Exploiting Residuals (2019, methodology): “[7] use outliers in point correspondences during ICP as seed for segmentation of dynamics.”
ConceptFusion: Open-set Multimodal 3D Mapping (2023, methodology): “3D feature fusion: We fuse f u,v,t into the global map following a 3D reconstruction pipeline [7].”
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].”
CNN-SLAM: Real-Time Dense Monocular SLAM with Learned Depth Prediction (2017, methodology): “From the point of view of the type of input data being processed, approaches can be classified into either depth camera-based [21, 30, 11] and monocular camera-based [22, 4, 20].”
Real-time 3D reconstruction at scale using voxel hashing (2013, background): “In this section we review relevant systems, with a focus on online reconstruction methods and active sensors.”
An Overview to Visual Odometry and Visual SLAM: Applications to Mobile Robotics (2015, methodology): “Another recent method proposed by Keller et al. [68] allows for 3D reconstruction of dynamic environments by automatically detecting dynamic changes in the scene.”
Gaussian Splatting SLAM (2023, positive): “Unlike other methods such as Elastic-Fusion [41] and PointFusion [9], however, by using differentiable rasterisation, our SLAM system can capture high-fidelity scene details and represent challenging object properties by direct optimisation against information from every pixel.”
Shading-based refinement on volumetric signed distance functions (2015, methodology): “Some hand-held RGB-D scanning approaches resort to point-based scene models [Keller et al. 2013; Weise et al. 2009].”