SVO: Fast semi-direct monocular visual odometry

By Christian Forster, Matia Pizzoli, Davide Scaramuzza. 2014

2,163 citations recorded by the source corpus.

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

RLBench: The Robot Learning Benchmark & Learning Environment (2019, neutral): “However, if we would like a manipulation system to make use of a SLAM map, it is not currently clear what the best way to represent this map is: whether it be sparse [54], [55], dense [56], [57], or semidense [58].”
Past, Present, and Future of Simultaneous Localization and Mapping: Toward the Robust-Perception Age (2016, neutral): “rement of dense method by exploiting only pixels with strong gradients (i.e., edges) [98, 113]; semi-direct methods instead leverage both sparse features (such as corners or edges) and direct methods [112, 113, 153] and are proven to be the most efﬁcient [113]; additionally, because they rely on spa”
ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual–Inertial, and Multimap SLAM (2020, methodology): “The hybrid system SVO [23], [24] extracts FAST features, uses a direct method to track features, and any pixel with nonzero intensity gradient, from frame to frame, and optimizes camera trajectory and 3D structure using reprojection error.”
Visual-Inertial Navigation for a Camera-Equipped 25g Nano-Quadrotor (2014, methodology): “the context of this paper – relatively large platforms, there are some exceptions: In [6], a visual SLAM system is used onboard a quadrotor weighing around 100 g.”
Gaussian Splatting SLAM (2023, neutral): “Dense SLAM: Dense visual SLAM focuses on reconstructing detailed 3D maps, unlike sparse SLAM methods which excel in pose estimation [4, 5, 21] but typically yield maps useful mainly for localisation.”
Robust visual inertial odometry using a direct EKF-based approach (2015, methodology): “By employing highly optimized SSEvectorized implementations, first real-time, CPU-based approaches for dense- or semi-dense motion estimation using a RGBD [15] or a monocular RGB camera [6], [7] have recently been proposed.”
On-Manifold Preintegration for Real-Time Visual--Inertial Odometry (2015, methodology): “3http://github.com/uzh-rpg/rpg svo 4http://borg.cc.gatech.edu 14 SVO [65] is a precise and robust monocular visual odometry system that employs sparse image alignment to estimate incremental motion and tracks features by minimizing the photometric error between subsequent frames.”
Automatic re-initialization and failure recovery for aggressive flight with a monocular vision-based quadrotor (2015, background): “…we start recovering when we measure ‖ã‖mean ‖cidle‖+ tlaunch, (29) where ‖ã‖mean is the norm of the measured acceleration averaged over the last 50 ms, cidle is the mass-normalized idle thrust that prevent the rotors from standing still and tlaunch = 2.0 m s−2 is a threshold parameter.”
Grape detection, segmentation and tracking using deep neural networks and three-dimensional association (2019, methodology): “A fast SLAM algorithm such as ORB-SLAM (Mur-Artal & Tardos, 2017) or SVO (Forster et al., 2014), not relying on any speciﬁc landmark, could be a more robust alternative.”
Direct Sparse Odometry (2016, methodology): “This is in contrast to hybrid approaches such as SVO [9], which revert to an indirect formulation for joint model optimization.”
Extending Monocular Visual Odometry to Stereo Camera Systems by Scale optimization (2019, background): “As an example, SVO [2] needs only about 3 milliseconds to process each frame.”
Probabilistic data association for semantic SLAM (2017, methodology): “The second optimization above is typically carried out via filtering [30]–[32] or pose-graph optimization [11], [12].”
Target following on nano-scale Unmanned Aerial Vehicles (2017, background): “For this first direction, the state-of-the-art is defined by high-level capabilities [7], [4] achieved on large (standard-size) robots operating in a power envelope of tens of Watts [8].”
REMODE: Probabilistic, monocular dense reconstruction in real time (2014, methodology): “At every time step k, the pose of the camera Tk,r in the depth map reference frame r is computed by a visual odometry routine that is based on recent advancement on semi-direct methods for camera localization [22].”
DynaSLAM II: Tightly-Coupled Multi-Object Tracking and SLAM (2020, background): “Most SLAM approaches assume a static scene and can only handle small fractions of dynamic content by labeling them as outliers to such static model [1]–[3].”
Neural RGB-D Surface Reconstruction (2021, background): “Reconstructing objects and scenes can be done using passive stereo systems that rely on stereo matching from two or multiple color views [29, 62], Structure-from-Motion [63], or SLAM-based [20, 21, 23] methods.”
A Machine Learning Approach to Visual Perception of Forest Trails for Mobile Robots (2016, methodology): “…shutter color camera (752 × 480 pixels), used for trail detection ; a down-looking MatrixVision mvBlueFox global shutter grayscale camera (752 × 480 pixels), used for feeding a Semidirect monocular Visual Odometry (SVO) pipeline [35]; an onboard Odroid-U3 system for image processing and control.”