Direct Sparse Odometry

By Jakob J. Engel, Vladlen Koltun, Daniel Cremers. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016

2,983 citations recorded by the source corpus.

Direct Sparse Odometry (DSO) is a visual odometry method based on a novel, highly accurate sparse and direct structure and motion formulation. It combines a fully direct probabilistic model (minimizing a photometric error) with consistent, joint optimization of all model parameters, including geometry-represented as inverse depth in a reference frame-and camera motion. This is achieved in real time by omitting the smoothness prior used in other direct methods and instead sampling pixels evenly throughout the images. Since our method does not depend on keypoint detectors or descriptors, it can naturally sample pixels from across all image regions that have intensity gradient, including edges or smooth intensity variations on essentially featureless walls. The proposed model integrates a full photometric calibration, accounting for exposure time, lens vignetting, and non-linear response functions. We thoroughly evaluate our method on three different datasets comprising several hours of video. The experiments show that the presented approach significantly outperforms state-of-the-art direct and indirect methods in a variety of real-world settings, both in terms of tracking accuracy and robustness.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/72e7f4edad44394d2bf63bb134707c8f991761c6

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

Citations in the corpus

Kimera: an Open-Source Library for Real-Time Metric-Semantic Localization and Mapping (2019, background): “First, while Back-end Geometry Sema-ntics ORB-SLAM [22] mono g2o points DSO [23] mono g2o points VINS-mono [24] mono/IMU Ceres points VINS-Fusion [25] mono/Stereo/IMU Ceres points ROVIOLI [26] stereo/IMU EKF points ElasticFusion [18] RGB-D alternation surfels Voxblox [27] RGB-D [26] TSDF SLAM++…”
MID-Fusion: Octree-based Object-Level Multi-Instance Dynamic SLAM (2018, background): “(sparse SLAM [4]) or an alternating optimisation (dense SLAM [5]) of map and pose relationship to solve SLAM.”
Geometry-Aware Learning of Maps for Camera Localization (2017, methodology): “Off-the-shelf VO algorithms [18, 19] provide relative camera poses between image pairs from these videos.”
D3VO: Deep Depth, Deep Pose and Deep Uncertainty for Monocular Visual Odometry (2020, methodology): “In [16], the residual is down-weighted when the pixels are with high image gradient to compensate small independent geometric noise [16].”
CubeSLAM: Monocular 3-D Object SLAM (2018, methodology): “Since there is no absolute scale for monocular DSO or ORB SLAM, we compute their pose error after scale alignment [15].”
VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator (2017, methodology): “Approaches that use only a monocular camera have gained significant interests in the field due to their small size, low-cost, and easy hardware setup [1]–[5].”
NICE-SLAM: Neural Implicit Scalable Encoding for SLAM (2021, methodology): “TANDEM [20] combines multi-view stereo with DSO [15] for a real-time dense SLAM system.”
Toward low-flying autonomous MAV trail navigation using deep neural networks for environmental awareness (2017, methodology): “And a third node runs the DSO visual odometry algorithm [8], whose output is converted to a camera-centric depth map for obstacle detection and avoidance.”
Efficient Surfel-Based SLAM using 3D Laser Range Data in Urban Environments (2018, background): “There has been significant advances in visionbased [6, 7] and RGB-D-based [18, 33, 3] SLAM systems over the past few years.”
PL-SLAM: A Stereo SLAM System Through the Combination of Points and Line Segments (2017, background): “intermediate representation, they are very sensitive to brightness changes (this phenomena was addressed in [11]) and constrained to narrow baseline motions.”
Visual SLAM algorithms: a survey from 2010 to 2016 (2017, methodology): “Recently, Engel et al. propose the direct sparse odometry (DSO) [53].”
DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras (2021, methodology): “One advantage of direct approaches is that they can model more information about the image, such as lines and intensity variations[15] which are not used by indirect approaches.”
A Comprehensive Survey of Visual SLAM Algorithms (2022, methodology): “Bottom: main steps followed by a direct method, that may result in a sparse (generated from the reconstruction of sequence_02 /TUM MonoVO [30] with the DSO algorithm [31]) or dense reconstruction (Reprinted from [25]), according to the chosen technique.”
OpenVSLAM: A Versatile Visual SLAM Framework (2019, methodology): “mera systems. UcoSLAM [9] adopts an algorithm that combines artificial landmarks, such as squared fiducial markers, and binary descriptor used by ORB–SLAM and ProSLAM. Meanwhile, LSD–SLAM [4] and DSO [3] are two different approaches of direct SLAM, which realizes visual SLAM processing directly expl”
Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art (2017, methodology): “Engel et al. (2014) extend their semi-dense method for visual odometry (Engel et al.”
Extending Monocular Visual Odometry to Stereo Camera Systems by Scale optimization (2019, methodology): “, [1]–[3]) are capable of camera tracking in these repetitive scenes and are computationally less expensive than multi-camera VO.”
MonST3R: A Simple Approach for Estimating Geometry in the Presence of Motion (2024, methodology): “Learning-based visual odometry replaces hand-designed parts of geometry-based methods (Mur-Artal et al., 2015; Mur-Artal & Tard´os, 2017; Engel et al., 2017) and enables large-scale training for better generalization even with moving objects.”
BAD SLAM: Bundle Adjusted Direct RGB-D SLAM (2019, methodology): “DSO has been extended with loop closure handling, but only using pose graph optimization without BA [20].”