A benchmark for RGB-D visual odometry, 3D reconstruction and SLAM

By Ankur Handa, Thomas Whelan, John McDonald Jr., Andrew J. Davison. 2014

937 citations recorded by the source corpus.

We introduce the Imperial College London and National University of Ireland Maynooth (ICL-NUIM) dataset for the evaluation of visual odometry, 3D reconstruction and SLAM algorithms that typically use RGB-D data. We present a collection of handheld RGB-D camera sequences within synthetically generated environments. RGB-D sequences with perfect ground truth poses are provided as well as a ground truth surface model that enables a method of quantitatively evaluating the final map or surface reconstruction accuracy. Care has been taken to simulate typically observed real-world artefacts in the synthetic imagery by modelling sensor noise in both RGB and depth data. While this dataset is useful for the evaluation of visual odometry and SLAM trajectory estimation, our main focus is on providing a method to benchmark the surface reconstruction accuracy which to date has been missing in the RGB-D community despite the plethora of ground truth RGB-D datasets available.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition, Geology, 3D Vision & Pose, Robotics & Control

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

In 3D Vision & Pose

In Robotics & Control

Citations in the corpus

Hypersim: A Photorealistic Synthetic Dataset for Holistic Indoor Scene Understanding (2020, background): “…the color image. alistic synthetic datasets and interactive simulation environments that have spurred rapid progress towards the goal of holistic indoor scene understanding [6, 7, 9, 10, 14, 15, 19, 22, 24, 31, 33, 36, 37, 39, 43, 44, 45, 46, 49, 52, 61, 62, 63, 65, 70, 72, 75, 76, 79, 83, 84].”
RGB-D datasets using microsoft kinect or similar sensors: a survey (2016, methodology): “20 Sample images of the office room scene taken at different camera poses from ICL-NUIM dataset [34]”
DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation (2019, methodology): “Following [25], we simulate the noise distribution of typical structure depth sensors, including Kinect V1 by adding zero-mean Guassian noise to the inverse depth representation of a ground truth input depth image:”
Playing for Data: Ground Truth from Computer Games (2016, methodology): “Synthetic scenes have been used for evaluating the robustness of image features [21] and for benchmarking the accuracy of visual odometry [16].”
Past, Present, and Future of Simultaneous Localization and Mapping: Toward the Robust-Perception Age (2016, neutral): “X X X [141, 239] X [134, 317, 318] X X X [122] X X X X X X X X [39] X X X X X [254] X X X X X Back-ends in small / medium scale scenarios [45, 186] X X Front-ends and back-ends in synthetic scenarios [135] X X X Loop-closures [124, 259] X X [213] X X [256] X X X X Long-term SLAM [25, 85] X X X [256]”
Playing for Benchmarks (2017, neutral): “Simulated worlds are commonly used to benchmark optical ﬂow algorithms [5, 4, 8] and visual odometry systems [22, 24, 64].”
Physically-Based Rendering for Indoor Scene Understanding Using Convolutional Neural Networks (2016, methodology): “[12, 11] introduced a laboriously created 3D scene dataset and demonstrated the usage on semantic segmentation training.”
ElasticFusion: Dense SLAM Without A Pose Graph (2015, neutral): “[7].”
InteriorNet: Mega-scale Multi-sensor Photo-realistic Indoor Scenes Dataset (2018, methodology): “We also support exporting the camera trajectory into other formats e.g. SLAM-Bench [19], SLAMBench2 [7], EuRoC [8], Freiburg [29] and ICL-NUIM [13], and also renderer formats such as OppositeRenderer (SceneNet RGB-D) [17].”
Implicit Neural Representations with Periodic Activation Functions (2020, background): “4, the proposed periodic activations significantly increase the details of objects and the complexity of scenes that can be represented by these neural SDFs, parameterizing a full room from the ICL-NUIM dataset [50] with only a single five-layer fully connected neural network.”
FutureMapping: The Computational Structure of Spatial AI Systems (2018, methodology): “The alternative, for this and other axes of evaluation, is to generate synthetic test data using computer graphics, such as the ICL-NUIM dataset [19].”
Tanks and temples (2017, background): “There are also benchmarks for related problems, such as mesh reconstruction from point clouds [Berger et al. 2013], visual odometry [Burri et al. 2016; Geiger et al. 2013], and RGB-D reconstruction [Choi et al. 2015; Handa et al. 2014; Sturm et al. 2012].”
Visual-Inertial Navigation Algorithm Development Using Photorealistic Camera Simulation in the Loop (2018, methodology): “Handa et al. provide a synthetic data set for the veriﬁcation of SLAM algorithms against a known 3D model and trajectory [15].”
SUN RGB-D: A RGB-D scene understanding benchmark suite (2015, neutral): “There are many existing RGB-D datasets available [54] [47] [1] [25] [44] [60] [49] [45] [29] [66] [57] [52] [46] [16] [21] [73] [35] [67] [3] [41] [10] [63] [42] [64] [65] [48] [12] [33] [8] [50] [26] [6].”
Deep Local Shapes: Learning Local SDF Priors for Detailed 3D Reconstruction (2020, methodology): “For synthetic scans, we use the ICL-NUIM RGBD benchmark dataset [23].”
Synthetic Data for Deep Learning (2019, methodology): “Handa et al. [233] provide their own simulated living room environment and dataset ICL-NUIM (Imperial College London and National University of Ireland Maynooth) with special emphasis on visual odometry and SLAM; they render high-quality RGBD images with ray tracing and take special care to model th”