Efficient & Effective Prioritized Matching for Large-Scale Image-Based Localization

By Torsten Sattler, Bastian Leibe, Leif Kobbelt. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017

656 citations recorded by the source corpus.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/0cdc707286a06e4fcfc0601515cedf714344772f

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

Understanding the Limitations of CNN-Based Absolute Camera Pose Regression (2019, methodology): “These matches are established by descriptor matching [21, 37, 39, 59, 63, 68, 69, 81] or by regressing 3D coordinates from pixel patches [8–10, 16, 23, 43, 46, 47, 65].”
InLoc: Indoor Visual Localization with Dense Matching and View Synthesis (2019, methodology): “Due to the sparse sampling of viewpoints in our indoor dataset, we cannot establish feature tracks between database images, preventing us from applying algorithms relying on co-visibility [23], [44], [64], [66].”
Comparative Evaluation of Hand-Crafted and Learned Local Features (2017, background): “In practice, feature matching is typically followed by a geometric verification stage to prune outliers [1, 17, 31, 34, 36, 37, 39].”
Learning Less is More - 6D Camera Localization via 3D Surface Regression (2017, methodology): “Compared to the PoseNet variants [10, 29] we improve accuracy by a factor of 10 for many scenes, and compared to the sparse feature-based Active Search [20] by a factor of 2.”
Global Pose Estimation with an Attention-Based Recurrent Network (2018, methodology): “RANSAC [13] is commonly used for geometric verification and outlier rejection, and there are also prioritized feature matching approaches [39].”
Visual Camera Re-Localization From RGB and RGB-D Images Using DSAC (2020, methodology): “ActiveSearch [7] utilizes a varying but limited receptive field.”
Egocentric Shopping Cart Localization (2018, methodology): “Another family of camera pose estimation methods leverage the 3D structure of the scene [4] [5] [24].”
Geometry-Aware Learning of Maps for Camera Localization (2017, methodology): “Examples include 3D landmarks for general visual SLAM methods [34, 35, 42], 3D lines and patches in semi-dense SLAM methods and indoor scenes [20, 43, 56], object-level context in semantic SLAM methods [10, 45], bag of visual word features on key frames for camera relocalization [12, 46, 47].”
Efficient Global 2D-3D Matching for Camera Localization in a Large-Scale 3D Map (2017, methodology): “To solve this scalability issue, existing direct 2D-3D methods often adopt advanced retrieval techniques, such as “vocabulary tree” and “ratio test” to remove ambiguous matches [40, 42–44 , 47 ].”
Privacy Preserving Image Queries for Camera Localization (2019, methodology): “Recent progress in image-based localization techniques have led to methods that are robust to changes in scene appearance and illumination [7, 57], scalable [36, 53, 54, 79], and efﬁcient [9, 15, 18, 28, 30, 36–38 , 69 ].”
From Coarse to Fine: Robust Hierarchical Localization at Large Scale (2018, methodology): “For Active Search [43], City Scale Localization [51], DenseVLAD [56], and NetVLAD [2], we use the evaluation reported in the paper introducing the benchmark.”
Image-Based Localization Using LSTMs for Structured Feature Correlation (2016, methodology): “Scene Area or Volume Active Search (w/o) [43] Active Search (w/) [43] PoseNet [23] Bayesian PoseNet [22] Proposed + Improvement(pos,ori)”
Privacy Preserving Image-Based Localization (2019, methodology): “Arguably, the most common approach to image-based localization is structure-based [19, 33, 46, 59] and tackles the problem by first matching the local 2D features of an image to the 3D point cloud model of the scene.”
Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art (2017, methodology): “Instead, more recent methods (Sattler et al. (2015, 2016)) use quantization into a fine vocabulary where each descriptor is represented by its word ID. Sattler et al. (2015) separate the difficult problem of finding a unique 2D-3D matching into two simpler ones.”
Geometric Loss Functions for Camera Pose Regression with Deep Learning (2017, methodology): “Area or Active Search PoseNet Bayesian PoseNet PoseNet (this work) PoseNet (this work) Scene Volume (SIFT) [43] (β weight) [22] PoseNet [19] Spatial LSTM [54] Learn σ(2) Weight Geometric Reprojection”
DUSt3R: Geometric 3D Vision Made Easy (2023, positive): “Our method obtains comparable accuracy compared to existing approaches, being feature-matching ones [101, 103] or end-to-end learning-based methods [11, 55, 102, 125, 152], even managing to out-perform strong baselines like HLoc [101] in some cases.”
Benchmarking 6DOF Outdoor Visual Localization in Changing Conditions (2018, methodology): “We evaluate Active Search [56] and the City-Scale Localization approach [67], a deterministic geometric outlier filter based on a known gravity direction, as representatives for efficient respectively scalable localization methods.”