Efficient Large-Scale Stereo Matching

By Andreas Geiger, Martin Roser, Raquel Urtasun. Asian Conference on Computer Vision, 2010

897 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/1c97127f828705328bceb6c5a50e2b1aefbb28ff

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Semi-global Stereo Matching with Surface Orientation Priors (2017, background): “For instance, surface priors can be derived from matched sparse feature points via triangulation [20] or plane fitting [42].”
DepthSynth: Real-Time Realistic Synthetic Data Generation from CAD Models for 2.5D Recognition (2017, background): “Since there is no theoretical upper bound on obtaining synthetically generated images to either train complex models for pose classiﬁcation [26,33,40,44] or ﬁll large databases for retrieval tasks [11, 42], research continues to gain impetus in this direction.”
Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches (2015, neutral): “43 67 2 Displets Güney and Geiger (2015) 2.”
Displets: Resolving stereo ambiguities using object knowledge (2015, background): “Local stereo methods [14, 20, 23, 27, 37, 46, 50, 67] often fail in challenging scenarios as they suffer from matching ambiguities in weakly textured, saturated or reflective regions.”
D Traffic Scene Understanding from Movable Platforms (2013, methodology): “We make use of the efficient large-scale stereo matching algorithm ELAS [32] to compute disparity maps for each frame in each sequence.”
3D Traffic Scene Understanding From Movable Platforms (2014, methodology): “We make use of the efficient large-scale stereo matching algorithm ELAS [32] to compute disparity maps for each frame in each sequence.”
Multi-scale object candidates for generic object tracking in street scenes (2016, methodology): “quence of stereo image pairs, we compute disparity maps using ELAS [23].”
Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art (2017, background): “Geiger et al. (2010) build a prior over the disparity space by forming a triangulation on a set of robustly matched correspondences, called support points.”
Nonparametric semantic segmentation for 3D street scenes (2013, methodology): “Specifically, we apply the efficient stereo matching algorithm proposed in [23] to compute a disparity map between stereo images, and then filter out extreme disparity values using a median filter with a 3×3 patch window for an image of resolution 1241×376.”
Vision-Based Offline-Online Perception Paradigm for Autonomous Driving (2015, background): “In this paper we focus on answering this question given the state-of-the-art in computer vision for semantic segmentation [32], 3D mapping [16], self-localization [21, 26] and object detection [13].”
Incremental dense semantic stereo fusion for large-scale semantic scene reconstruction (2015, methodology): “2) as a result, the estimated depth [25] is usually more noisy; 3) the depth range is much larger and 4) dynamically moving objects are much more common and the camera itself may move significantly between consecutive frames (e.”
Hierarchical Deep Stereo Matching on High-Resolution Images (2019, methodology): “0 on all pixels increases by 10.3% while being 1.46x faster, which is still more accurate than ELAS, SGBM and iResNet.”
SDC – Stacked Dilated Convolution: A Unified Descriptor Network for Dense Matching Tasks (2019, methodology): “From top to bottom: Disparity map for ELAS [12] on ETH3D [30], optical ﬂow for CPM [19] on Sintel [6], and scene ﬂow (disparity and optical ﬂow components) for SFF [33] on KITTI [26]. on best practice design decisions from other domains which results in the use of pooling or other striding layers.”
StereoScan: Dense 3d reconstruction in real-time (2011, methodology): “Our real-time parameterization uses the standard settings for LIBELAS stereo matching [12], and sparse feature matching at 2 scales with 50 RANSAC iterations , an inlier threshold of 1.5 pixels and τ disp = τ f low = 5 pixels.”
Are we ready for autonomous driving? The KITTI vision benchmark suite (2012, neutral): “For stereo matching, we run global [26, 37, 46], semiglobal [23], local [5, 20, 38] and seed-growing [27, 10, 9] methods.”
Robust stereo matching with surface normal prediction (2017, background): “These algorithms [21], [19], [13], [38] generally compute the disparity for each pixel independently, so the recovered disparity maps may easily have artifacts around textureless and occluded regions.”
Vision Enhancement in Homogeneous and Heterogeneous Fog (2012, methodology): “The point in using these images is that the disparity maps obtained using the stereo reconstruction algorithm Libelas [19] 2www.lcpc.fr/english/products/image-databases/ article/frida-foggy-road-image-database 3http://www.rainsoft.de/software/datasets.html ha l-0 07 07 03 9, v er si on 1 - 12 J un 2”
Multimodal information fusion for urban scene understanding (2016, methodology): “bThe disparitymap is computed from the ELAS algorithm [16].”