Highly scalable appearance-only SLAM - FAB-MAP 2.0

By Mark Joseph Cummins, Paul M. Newman. Robotics: Science and Systems Conference, 2009

364 citations recorded by the source corpus.

We describe a new formulation of appearance-only SLAM suitable for very large scale navigation. The system navigates in the space of appearance, assigning each new observation to either a new or previously visited location, without reference to metric position. The system is demonstrated performing reliable online appearance mapping and loop closure detection over a 1,000 km trajectory, with mean filter update times of 14 ms. The 1,000 km experiment is more than an order of magnitude larger than any previously reported result. The scalability of the system is achieved by defining a sparse approximation to the FAB-MAP model suitable for implementation using an inverted index. Our formulation of the problem is fully probabilistic and naturally incorporates robustness against perceptual aliasing. The 1,000 km data set comprising almost a terabyte of omni-directional and stereo imagery is available for use, and we hope that it will serve as a benchmark for future systems.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition, Ocean Engineering

View Paper: https://www.semanticscholar.org/paper/db39b7be3c56071b610f5557aa93066d442d529c

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 Ocean Engineering

Citations in the corpus

Past, Present, and Future of Simultaneous Localization and Mapping: Toward the Robust-Perception Age (2016, neutral): “ace and allowing more efﬁcient search. Bag-of-words can be arranged into hierarchical vocabulary trees [231] that enable efﬁcient lookup in large-scale datasets. Bag-of-words-based techniques such as [79, 80, 121] have shown reliable performance on the task of single session loop closure detection. ”
SeqSLAM: Visual route-based navigation for sunny summer days and stormy winter nights (2012, methodology): “isual recognition of places and place sequences in the field of robotic navigation has advanced rapidly in recent years, with several techniques [1-5] now being competitive with laser range-based techniques.”
Learning and Calibrating Per-Location Classifiers for Visual Place Recognition (2013, background): “Similar to other work in large scale place recognition [7, 13, 27] and image retrieval [20, 21, 28], we build on the bag-of-visual-words representation [6, 28] and describe each image by a set of quantized local invariant features, such as SURF [1] or SIFT [17].”
Appearance-only SLAM at large scale with FAB-MAP 2.0 (2011, background): “The material was first presented in [13]; here we expand the presentation with a more detailed treatment and additional results.”
24/7 Place Recognition by View Synthesis (2018, neutral): “The large-scale place recognition is often formulated as a variation of image retrieval [22, 33] where the query photograph is localized by matching it to a large database of geo-tagged images such as Google street-view [6, 8, 10, 14, 24, 35, 36, 40, 44].”
Get Out of My Lab: Large-scale, Real-Time Visual-Inertial Localization (2015, methodology): “A loopclosure pipeline, combining the work of Galvez-Lopez and Tardos [16] with the FABMAP algorithm by Cummins and Newman [9], identifies places that have been visited multiple times.”
Scale Drift-Aware Large Scale Monocular SLAM (2010, background): “It is well known that loop closures can be detected effectively using appearance information only [1, 5].”
Visual SLAM: Why filter? (2012, neutral): “Furthermore, it was shown recently that loop-closing can be solved efficiently using appearance-based methods [41, 8] which can be formulated independently from metric SLAM systems.”
Visual Place Recognition with Repetitive Structures (2015, background): “Scalable place recognition methods [3, 7, 18, 31, 37] often build on the efficient bag-of-visual-words representation developed for object and image retrieval [6, 13, 15, 24, 26, 40].”
Visual Place Recognition: A Survey (2016, methodology): “FAB-MAP 2.0 [87], [142] also used an inverted index with a bag-of-words model to demonstrate visual place recognition along a 1000-km path.”
Appearance-Based Loop Closure Detection for Online Large-Scale and Long-Term Operation (2013, background): “When p ( S t | L t ) has been updated and normalized, the highest loop closure hypothesis S t = i of p ( S t | L t ) is accepted if the new location hypothesis p ( S t = − 1 | L t ) is lower than the loop closure threshold T loop (set between 0 and 1).”
Deep learning features at scale for visual place recognition (2017, methodology): “We tested using four benchmark place recognition datasets, with details summarized in Table I.”
NetVLAD: CNN Architecture for Weakly Supervised Place Recognition (2015, background): “While there have been many improvements in designing better image retrieval [2, 3, 11, 12, 17, 25, 26, 27, 29, 32, 47, 50, 51, 52, 53, 70, 77, 78, 81] and place recognition [4, 9, 10, 15, 16, 24, 35, 43, 45, 62, 63, 64, 74, 79, 80] systems, not many works have performed learning for these tasks.”
Double window optimisation for constant time visual SLAM (2011, background): “Candidates for large loop closures can be efficiently detected using appearance information only [2, 14].”
Geometric Loss Functions for Camera Pose Regression with Deep Learning (2017, methodology): “Traditionally, this has been modelled as an image retrieval problem [6, 9, 54, 46] enabling the use of efficient and scalable retrieval approaches [35, 38] such as Bag-of-Words (BoW) [48], VLAD [17, 10], and Fisher vectors [18].”
The three-dimensional normal-distributions transform : an efficient representation for registration, surface analysis, and loop detection (2009, methodology): “) Though the chosen term for the problem is “loop detection” in this text, the proposed method solves the same problem that Cummins and Newman [25] refer to as “appearance-only SLAM”.”
Scalable Nearest Neighbor Algorithms for High Dimensional Data (2014, background): “, [60], [61], [62], [63], [64]) and is widely used in the computer vision community, in part due to its inclusion in OpenCV [65], the popular open source computer vision library.”
Practice makes perfect? Managing and leveraging visual experiences for lifelong navigation (2012, background): “Metric trajectory estimation systems using laser [8] and vision [9], [10] have been demonstrated while large scale topological loop-closure systems using vision have been achieved [11] and crude GPS with a <10m error is often available in outdoor environments.”