Semi-dense Visual Odometry for a Monocular Camera

By Jakob J. Engel, Jürgen Sturm, Daniel Cremers. IEEE International Conference on Computer Vision, 2013

578 citations recorded by the source corpus.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/bfab61cbdf0120514ef034eaf08df8e440e310e2

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

Deep Virtual Stereo Odometry: Leveraging Deep Depth Prediction for Monocular Direct Sparse Odometry (2018, methodology): “CNN-SLAM [39] extends LSD-SLAM [9] by predicting depth with a CNN and reﬁning the depth maps using Bayesian ﬁltering [9,7].”
PL-SVO: Semi-direct Monocular Visual Odometry by combining points and line segments (2016, background): “For dealing with the high computational requirements of direct methods, a novel monocular technique is proposed in [21], where authors estimate the camera motion in a semi-dense approach, thus reaching real-time performance on a CPU.”
Stereo DSO: Large-Scale Direct Sparse Visual Odometry with Stereo Cameras (2017, methodology): “Each time a new stereo frame is fed into the system, direct image alignement [10] is used to track it with respect to the newest keyframe in the active window: a constant motion model is used to assign an initial pose to the new frame to be tracked.”
Geometry-Aware Learning of Maps for Camera Localization (2017, methodology): “We can compute the relative poses v̂ij between consecutive frames with visual odometry algorithms [18, 19, 44].”
Visual SLAM algorithms: a survey from 2010 to 2016 (2017, methodology): “The relationship between vSLAM and VO can also be found from the papers [20, 21] and the papers [22, 23].”
Direct visual-inertial odometry with stereo cameras (2016, background): “In turn, the effect of noisy poses is approximated during depth estimation [7].”
Pop-up SLAM: Semantic monocular plane SLAM for low-texture environments (2016, methodology): “(2) Otherwise, if a pixel has a propagated depth dl with variance σ(2) l from LSD, we fuse it with the pop-up depth dp of variance σ(2) p using the filtering approach [23]:”
SVO: Semidirect Visual Odometry for Monocular and Multicamera Systems (2017, methodology): “When a depth ﬁlter’s uncertainty becomes small enough, a new 3D point is inserted in the map and is immediately used for motion estimation.”
LSD-SLAM: Large-Scale Direct Monocular SLAM (2014, methodology): “For comparison we show respective results from semi-dense mono-VO [9], keypoint-based mono-SLAM [15], direct RGB-D SLAM [14] and keypoint-based RGB-D SLAM [7].”
Large-scale direct SLAM for omnidirectional cameras (2015, methodology): “We then perform line-stereo the same way as is done in [22].”
Real-Time 3D Reconstruction and 6-DoF Tracking with an Event Camera (2016, methodology): “We penalise deviation from a spatially smooth inverse depth map by assigning each inverse depth value the average of its neighbours, weighted by their respective inverse variances as described in [9].”
D3VO: Deep Depth, Deep Pose and Deep Uncertainty for Monocular Visual Odometry (2020, methodology): “Unlike traditional direct VO approaches [19,23] which initialize the front-end tracking for each new frame with a constant velocity motion model, we leverage the predicted poses between consecutive frames to build a non-linear factor graph [41, 47].”
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.”
SVO: Fast semi-direct monocular visual odometry (2014, methodology): “To reduce the computational demand, the method described in [23], which was published during the review process of this work, uses only pixels characterized by strong gradient.”
Large-scale direct SLAM with stereo cameras (2015, methodology): “After subpixel accurate refinement of the disparity, its variance is estimated through the geometric and photometric error identified in [16].”
Introspective perception: Learning to predict failures in vision systems (2016, background): “This application is particularly interesting given the wide spectrum of challenges involved in fault tolerance control in perception for autonomous aerial systems navigating outdoor diverse environments based on a single sensor [32].”
DeepVO: Towards end-to-end visual odometry with deep Recurrent Convolutional Neural Networks (2017, background): “Recently, semi-direct approaches which realise superior performance are developed for the monocular VO [12]– [14].”
Pose Estimation for Augmented Reality: A Hands-On Survey (2016, background): “They are able to provide precise models, which can later be used in markerless model-based localization methods (see Section 3.2).”
CNN-SLAM: Real-Time Dense Monocular SLAM with Learned Depth Prediction (2017, methodology): “every element of the key-frame is refined, in contrast to [5] that only refines depth values along high gradient regions.”