Pith. sign in

REVIEW 3 major objections 5 minor 31 references

Continuous Map Matching to Paths under Travel Time Constraints

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Continuous map matching to paths is guaranteed to find a feasible mapping whenever one exists.

desk verdict The continuous interval-sweep algorithm for map matching on path graphs is a real step forward, but the subquadratic running-time lemma has a proof gap that needs fixing before the main theoretical claim is solid. read the letter →

arxiv 2506.18354 v1 pith:7R2DOD4J submitted 2025-06-23 cs.CG cs.DS

Yannick Bosch , Sabine Storandt This is my paper · ORCID
classification cs.CGcs.DS MSC 68U0568W40
keywords mapmatchingtraveltimeconstraintspathgraphscontinuouscandidatelocationsintervalsweepsegment-circleintersectioncomputationalgeometrypublictransitdata
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Map matching normally snaps each measurement to a shortlist of candidate points on a road network, then checks whether travel times between consecutive candidates fit the recorded timestamps. On a path graph, this paper shows, that discretization can miss a feasible match entirely: the only valid contact point may lie in the interior of a road segment, invisible to any endpoint-based candidate list. The authors introduce COMMA, which treats the whole intersection of each measurement's uncertainty disk with the path as the candidate set, represented compactly as intervals along the path. Two sweeps, one forward and one backward, trim these intervals by the travel-time constraints, and the paper proves that surviving intervals are exactly the locations that participate in some feasible mapping. The result is a guarantee the classical DAG approach lacks, together with running times of $\mathcal{O}(k^2 n \log n k)$ in general and $\mathcal{O}(k n^{0.695} + n \log^3 n)$ under realistic sparsity assumptions on the disks.

What carries the argument

The machinery is the interval representation of continuous candidate sets, sharpened by a segment-circle intersection data structure. Every measurement's candidate region $D_i\cap P$ is a sorted union of intervals with at most two boundary crossings per path edge, so the infinite set of possible snap locations is encoded by $\mathcal{O}(n)$ interval endpoints. The forward sweep trims right endpoints using the largest reachable endpoint from the previous layer, the backward sweep trims left endpoints using the earliest start of the next layer, and the surviving intervals function as a certificate of feasibility. To make the interval computation fast, the paper separates segment-circle intersections into two types: one endpoint inside the circle (reported by a balanced tree over the path with convex-hull tests in $\mathbb{R}^3$) and both endpoints outside (reported by a triangle-enclosure data structure on a truncated strip around each segment). For segments shorter than the circle diameter, the paper contributes a new two-triangle construction that covers the relevant strip while excluding the overlap $D_x\cap D_y$ of the two endpoint disks, which is what keeps the type-(ii) reporting exact.

What would settle it

Construct a path whose segments are clustered so that every measurement disk overlaps every other disk and each disk boundary crosses $\Theta(n)$ segments, then measure COMMA's scaling in $n$ and $k$; if the time follows $k^2 n \log n k$ rather than $k n^{0.695} + n \log^3 n$, the sparsity assumptions behind Theorem 7 are violated. Separately, take a segment shorter than the circle diameter and query the type-(ii) data structure with a circle centered in $D_x \cap D_y$; the paper's two-triangle construction should report no intersection, and a direct implementation check would confirm or refute that coverage claim.

Watch

Extended reading notes

Core claim

The central claim is that feasibility for travel-time-constrained map matching on a path graph can be decided without ever choosing discrete candidate locations. For each measurement $M_i$, the acceptable locations are the points of the path $P$ inside the disk $D_i$ of radius $r$, and these form a union of at most $n$ intervals. COMMA computes those intervals, then propagates reachability in two sweeps: the forward sweep caps each interval's right end at the farthest point that can be reached from some interval of the previous measurement within the timestamp difference, and the backward sweep raises each interval's left end to the earliest point that can still reach some interval of the next measurement. Lemma 1 shows the surviving intervals are exactly the locations that lie on some feasible mapping sequence, so any nonempty final interval set yields a feasible match and any empty set certifies infeasibility. Lemma 3 bounds the general running time by $\mathcal{O}(k^2 n \log n k)$, and Theorem 7 improves this to $\mathcal{O}(k n^{0.695} + n \log^3 n + k\sigma\log^2 n)$ under the realistic model in which each disk overlaps only a constant number of other disks and each disk boundary cuts at most $\sigma$ path segments.

Load-bearing premise

The fast $\mathcal{O}(k n^{0.695} + n \log^3 n)$ running time rests on the assumption that each measurement disk overlaps only a constant number of other disks and that each disk boundary intersects at most $\sigma$ path segments; if real inputs do not satisfy this, only the slower $\mathcal{O}(k^2 n \log n k)$ bound is proven.

Editorial extensions

If this is right

  • For path-shaped networks such as bus lines, COMMA replaces sampling-based candidate snapping with an exact feasibility test: either it returns a consistent mapping or it certifies that none exists.
  • Because the interval sets encode every feasible mapping rather than one snapshot, secondary objectives such as minimizing the summed distance between measurements and matched locations can be optimized over the surviving intervals without losing the feasibility guarantee.
  • Under the realistic sparsity model, the running time is subquadratic in $n$ even when the number of measurements $k$ grows linearly with $n$, while the DAG baseline is quadratic and may still output no solution.
  • The segment-circle reporting data structure, including the short-segment triangle construction, is a self-contained geometric tool that applies to any batch of fixed-radius circle-versus-segment intersection queries, not only to map matching.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same interval-sweep logic suggests a route to general graphs: enumerate the points where disk boundaries cross edges, then ask which of those crossing points can reach the next measurement's crossings within the time budget; the paper leaves this extension open, and the continuous-candidate insight indicates why endpoint sampling fails on general networks as well.
  • One testable prediction is that COMMA's practical speedup should shrink on adversarial inputs where many disks overlap and each boundary cuts many segments, degrading toward the $\mathcal{O}(k^2 n \log n k)$ worst case; experiments on such inputs would delimit when the realistic model is safe to assume.
  • If the interval representation is embedded as a subroutine in candidate-path approaches for general-graph map matching, the per-path feasibility check becomes exact, which could remove the sampling bias that current multi-step matchers inherit from discrete candidate sets.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper considers continuous map matching on an embedded path graph with travel time constraints: each spatio-temporal measurement may be mapped to any point in a disk of fixed radius around it, and consecutive mapped points must be reachable along the path within the timestamp difference. The authors argue that standard DAG-based algorithms with finite candidate sets can miss feasible solutions, and they propose COMMA, which represents all feasible mapping locations as unions of intervals on the path and updates these intervals by forward and backward sweeps. They claim that COMMA always finds a feasible mapping if one exists, with running time O(k^2 n log nk) in general and an improved O(k n^0.695 + n log^3 n) bound under a sparsity model. The improved bound relies on new segment-circle intersection data structures, one tailored to paths and one adapted from known triangle-enclosure structures. Experiments on synthetic data and GTFS bus-stop data show large practical speedups over a DAG baseline with sampled candidates.

Significance. If the main claims are correct, this is a genuinely useful contribution: it replaces an inherently incomplete discrete-candidate approach with a continuous one that is guaranteed to find a feasible matching when one exists, at least for path graphs. The path-specific type-(i) intersection data structure in Section 4.1.1 is elegant, and the experimental comparison, especially on GTFS data, is practically relevant. The paper is also careful to state its sparsity assumptions explicitly. However, the two principal theoretical results — the general correctness argument for overlapping disks and the improved running-time theorem — rest on arguments that are incomplete or, in the case of the type-(ii) triangle construction, not exact as stated. The core idea is promising and probably repairable, but the manuscript needs substantial revision before the central claims can be accepted.

major comments (3)
  1. [Section 3.2] The correctness proof for the overlapping-disk case is not supplied. Lemma 1 is proved under the implicit assumption that intervals in consecutive layers are disjoint: the argument uses 'the interval with smallest starting point greater than b' as the only possible successor interval. Once disks overlap, an interval of I_{i+1} may start inside [a,b], and the backward sweep must subdivide intervals at such starts/ends; the paper only says 'the backward sweep works analogue' without defining the analogue or proving an invariant. Since the claim that COMMA 'always detects a consistent map matching path (if one exists)' includes the overlapping case, this is a load-bearing gap. Please add a formal invariant for the subdivided intervals and a correctness proof covering both sweeps under overlap.
  2. [Section 4.1.2, Lemma 6] The alternative triangle construction for segments shorter than 2r over-covers the region R_s \ (D_x ∪ D_y), so the type-(ii) data structure reports segments that intersect the query circle exactly once. For example, put s along the x-axis from (0,0) to (L,0) with L < 2r and consider the upper triangle with base at y=r and apex at the upper intersection of the two disks. A point such as (ε, r−δ) with small ε,δ lies inside this triangle and inside D_x but outside D_y; a query circle centered there has exactly one intersection with s. Consequently Lemma 6's assertion that the structure reports the q segments intersecting the circle is false as stated, and the proof of Theorem 7 inherits the problem. The union of the type-(i) and type-(ii) outputs may still be correct after deduplication, and the extra reports are legitimate intersecting segments, so the asymptotic bound may be recoverable by a constant-factor output argument; but the exact-reporting lemma and the proof of Theorem 7 need to be revised explicitly.
  3. [Lemma 3 and Section 3.2] The running-time analysis for the overlapping case is asserted rather than fully proved. The claim that 'the number of interval access operations is bounded by |I_{i-1}| + |I_i|' depends on a charging argument over interval overlaps that is not written down, and the statement that 'there can be at most kn interval starting points per disk' needs a formal accounting for both the forward sweep, the backward sweep, and the interval merging step. Because this bound is the basis of the O(k^2 n log nk) worst-case claim, please replace the informal paragraph with a precise invariant and a charging argument.
minor comments (5)
  1. [Lemma 4] The claim that the intersection of the query plane with a three-dimensional convex hull can be checked in O(log n) time is not self-contained; please provide a reference to a standard linear-programming or convex-polytope query result, or describe the oracle.
  2. [Section 4 and Theorem 7] The sparsity model assumes both that each disk overlaps only a constant number of disks and that each disk boundary intersects at most sigma segments, but the proof of Theorem 7 appears to use only the sigma bound on segment intersections. Please clarify whether the constant-overlap assumption is actually needed, and if so, where.
  3. [Section 5.2] The experiments use a CGAL AABB tree instead of the theoretically analyzed data structures, so the practical results do not validate the O(n^0.695)-type bound. Please state this limitation explicitly in the experimental section.
  4. [Section 4.1.1] In the original construction of [17], the s-tree is built on a spanning path over the segment endpoints, while Lemma 4 builds it directly on the path segments. Please clarify the relationship and confirm that the leaf-level convex hull criterion detects exactly the type-(i) segments.
  5. [Throughout] The symbol q is used both for output size in the data-structure lemmas and for the number of reported segments in Theorem 7; please define the output-size convention in each lemma, especially whether duplicates are counted.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the algorithm's correctness is proved from the sweep invariants, and running-time bounds rely on external data-structure results, not on the paper's own assumptions.

full rationale

The paper's central claim, that COMMA always finds a feasible matching sequence when one exists, is supported by a direct proof (Lemma 1) about the intervals produced by the forward and backward sweeps. The sweep rules are defined purely in terms of predecessor/successor intervals and travel-time budgets; the existence of a feasible successor is proved, not assumed. The running-time claims are likewise not circular: the O(k^2 n log nk) bound follows from an explicit counting argument over interval starting points, and the improved O(k n^0.695 + n log^3 n) bound depends on the external segment-circle intersection data structures of Gupta–Janardan–Smid [17] and Overmars–Schipper–Sharir [25]. None of these results is imported from the authors' own prior work in a load-bearing way. The only self-citations, e.g., [11] for the DAG baseline and [12] for trajectory indexing, appear in related work or experimental context and do not justify the correctness or complexity theorems. The geometric triangle-coverage assertion in Section 4.1.2 may be debatable as a correctness matter, but it is not circular: the data structure is intended to report intersecting segments, and the algorithm's feasibility guarantee does not define its output in terms of that report. There are no fitted constants, no predictions that reduce to inputs, and no renamed known results at the core of the derivation. The derivation is self-contained against external benchmarks, so the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The correctness of the basic sweep needs only the first domain axiom; the running-time claims additionally import two black-box data structures from the literature and an ad hoc sparsity model. There are no fitted free parameters and no invented entities.

assumptions (4)
  • domain assumption Travel time along the directed path P is additive and positive, so the time from a point at parameter x to a point at parameter y is y - x for y >= x.
    Used throughout Section 3 to reduce feasibility to ordered intervals on a line; back-tracking or bidirectional edges would invalidate the interval representation.
  • standard math The lifted-point transformation of Gupta, Janardan and Smid [17] maps each query circle to a plane such that a point lies inside the circle iff its lifted image lies below the plane.
    Adopted as a black box in Section 4.1.1; it underpins the type (i) intersection reporting structure.
  • standard math The Overmars-Schipper-Sharir partition tree [25] reports triangle enclosures in O(n^0.695 + q) time after O(n log^3 n) preprocessing.
    Used in Lemma 6 for the improved running time; no implementation is provided or checked.
  • ad hoc to paper In realistic inputs, each disk overlaps only a constant number of other disks and each disk boundary is intersected by at most sigma path segments.
    Assumed in Section 4 to obtain the O(k n^0.695 + n log^3 n) bound; not a consequence of the problem formulation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Continuous Map Matching to Paths under Travel Time Constraints." pith.science (2026). https://pith.science/paper/7R2DOD4J

@misc{pith2026250618354,
  author       = {Pith},
  title        = {Pith review of: Continuous Map Matching to Paths under Travel Time Constraints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7R2DOD4J}},
  note         = {Machine review of arXiv:2506.18354}
}
abstract

In this paper, we study the problem of map matching with travel time constraints. Given a sequence of $k$ spatio-temporal measurements and an embedded path graph with travel time costs, the goal is to snap each measurement to a close-by location in the graph, such that consecutive locations can be reached from one another along the path within the timestamp difference of the respective measurements. This problem arises in public transit data processing as well as in map matching of movement trajectories to general graphs. We show that the classical approach for this problem, which relies on selecting a finite set of candidate locations in the graph for each measurement, cannot guarantee to find a consistent solution. We propose a new algorithm that can deal with an infinite set of candidate locations per measurement. We prove that our algorithm always detects a consistent map matching path (if one exists). Despite the enlarged candidate set, we also demonstrate that our algorithm has superior running time in theory and practice. For a path graph with $n$ nodes, we show that our algorithm runs in $\mathcal{O}(k^2 n \log {nk})$ and under mild assumptions in $\mathcal{O}(k n ^\lambda + n \log^3 n)$ for $\lambda \approx 0.695$. This is a significant improvement over the baseline, which runs in $\mathcal{O}(k n^2)$ and which might not even identify a correct solution. The performance of our algorithm hinges on an efficient segment-circle intersection data structure. We describe how to design and implement such a data structure for our application. In the experimental evaluation, we demonstrate the usefulness of our novel algorithm on a diverse set of generated measurements as well as GTFS data.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages

  1. [1]

    Sparse map-matching in public transit networks with turn restrictions

    Hannah Bast and Patrick Brosi. Sparse map-matching in public transit networks with turn restrictions. In Proceedings of the 26th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems , pages 480--483, 2018

    work page 2018

  2. [2]

    An introduction to map matching for personal navigation assistants

    David Bernstein, Alain Kornhauser, et al. An introduction to map matching for personal navigation assistants. 1996

    work page 1996

  3. [3]

    On map-matching vehicle tracking data

    Sotiris Brakatsoulas, Dieter Pfoser, Randall Salas, and Carola Wenk. On map-matching vehicle tracking data. In Proceedings of the 31st international conference on Very large data bases , pages 853--864, 2005

    work page 2005

  4. [4]

    Map-matching using shortest paths

    Erin Chambers, Brittany Terese Fasy, Yusu Wang, and Carola Wenk. Map-matching using shortest paths. ACM Transactions on Spatial Algorithms and Systems (TSAS) , 6(1):1--17, 2020

    work page 2020

  5. [5]

    Quasi-optimal range searching in spaces of finite vc-dimension

    Bernard Chazelle and Emo Welzl. Quasi-optimal range searching in spaces of finite vc-dimension. Discrete & Computational Geometry , 4:467--489, 1989

    work page 1989

  6. [6]

    Approximate map matching with respect to the fr \'e chet distance

    Daniel Chen, Anne Driemel, Leonidas J Guibas, Andy Nguyen, and Carola Wenk. Approximate map matching with respect to the fr \'e chet distance. In 2011 Proceedings of the Thirteenth Workshop on Algorithm Engineering and Experiments (ALENEX) , pages 75--83. SIAM, 2011

    work page 2011

  7. [7]

    Algorithms for ray-shooting and intersection searching

    Siu Wing Cheng and Ravi Janardan. Algorithms for ray-shooting and intersection searching. Journal of Algorithms , 13(4):670--692, 1992

    work page 1992

  8. [8]

    A basis of spatial big data analysis with map-matching system

    Wonhee Cho and Eunmi Choi. A basis of spatial big data analysis with map-matching system. Cluster Computing , 20(3):2177--2192, 2017

    work page 2017

Show all 31 references
  1. [9]

    Physically consistent map matching

    Bram Custers, Wouter Meulemans, Marcel Roeloffzen, Bettina Speckmann, and Kevin Verbeek. Physically consistent map matching. In Proceedings of the 30th International Conference on Advances in Geographic Information Systems , pages 1--4, 2022

    2022

  2. [10]

    Path segmentation for beginners: an overview of current methods for detecting changes in animal movement patterns

    Hendrik Edelhoff, Johannes Signer, and Niko Balkenhol. Path segmentation for beginners: an overview of current methods for detecting changes in animal movement patterns. Movement ecology , 4:1--21, 2016

    2016

  3. [11]

    Algorithms for matching and predicting trajectories

    Jochen Eisner, Stefan Funke, Andre Herbst, Andreas Spillner, and Sabine Storandt. Algorithms for matching and predicting trajectories. In 2011 Proceedings of the Thirteenth Workshop on Algorithm Engineering and Experiments (ALENEX) , pages 84--95. SIAM, 2011

    2011

  4. [12]

    Pathfinder: storage and indexing of massive trajectory sets

    Stefan Funke, Tobias Rupp, Andr \'e Nusser, and Sabine Storandt. Pathfinder: storage and indexing of massive trajectory sets. In Proceedings of the 16th International Symposium on Spatial and Temporal Databases , pages 90--99, 2019

    2019

  5. [13]

    Snapping stops: Matching stops to the network, n.d

    geOps . Snapping stops: Matching stops to the network, n.d. Accessed: 2025-01-03. URL: https://geops.com/de/blog/snapping-stops

    2025

  6. [14]

    A novel map-matching procedure for low-sampling gps data with applications to traffic flow analysis

    Luca Giovannini. A novel map-matching procedure for low-sampling gps data with applications to traffic flow analysis. 2011

    2011

  7. [15]

    Online map-matching based on hidden markov model for real-time traffic sensing applications

    Chong Yang Goh, Justin Dauwels, Nikola Mitrovic, Muhammad Tayyab Asif, Ali Oran, and Patrick Jaillet. Online map-matching based on hidden markov model for real-time traffic sensing applications. In 2012 15th International IEEE Conference on Intelligent Transportation Systems ,...

    2012

  8. [16]

    Map matching queries on realistic input graphs under the fr \'e chet distance

    Joachim Gudmundsson, Martin P Seybold, and Sampson Wong. Map matching queries on realistic input graphs under the fr \'e chet distance. ACM Transactions on Algorithms , 20(2):1--33, 2024

    2024

  9. [17]

    On intersection searching problems involving curved objects

    Prosenjit Gupta, Ravi Janardan, and Michiel Smid. On intersection searching problems involving curved objects. In Algorithm Theory—SWAT'94: 4th Scandinavian Workshop on Algorithm Theory Aarhus, Denmark, July 6--8, 1994 Proceedings 4 , pages 183--194. Springer, 1994

    1994

  10. [18]

    A machine learning approach to improve the accuracy of gps-based map-matching algorithms

    Mahdi Hashemi and Hassan A Karimi. A machine learning approach to improve the accuracy of gps-based map-matching algorithms. In 2016 IEEE 17th International Conference on Information Reuse and Integration (IRI) , pages 77--86. IEEE, 2016

    2016

  11. [19]

    If-matching: Towards accurate map-matching with information fusion

    Gang Hu, Jie Shao, Fenglin Liu, Yuan Wang, and Heng Tao Shen. If-matching: Towards accurate map-matching with information fusion. IEEE Transactions on Knowledge and Data Engineering , 29(1):114--127, 2016

    2016

  12. [20]

    Survey on vehicle map matching techniques

    Zhenfeng Huang, Shaojie Qiao, Nan Han, Chang-an Yuan, Xuejiang Song, and Yueqiang Xiao. Survey on vehicle map matching techniques. CAAI Transactions on Intelligence Technology , 6(1):55--71, 2021

    2021

  13. [21]

    Map-matched trajectory compression

    Georgios Kellaris, Nikos Pelekis, and Yannis Theodoridis. Map-matched trajectory compression. Journal of Systems and Software , 86(6):1566--1579, 2013

    2013

  14. [22]

    Map matching with travel time constraints

    John Krumm, Eric Horvitz, and Julie Letchner. Map matching with travel time constraints. Technical report, SAE Technical Paper, 2007

    2007

  15. [23]

    Map-matching for low-sampling-rate gps trajectories

    Yin Lou, Chengyang Zhang, Yu Zheng, Xing Xie, Wei Wang, and Yan Huang. Map-matching for low-sampling-rate gps trajectories. In Proceedings of the 17th ACM SIGSPATIAL international conference on advances in geographic information systems , pages 352--361, 2009

    2009

  16. [24]

    Accurate real-time map matching for challenging environments

    Reham Mohamed, Heba Aly, and Moustafa Youssef. Accurate real-time map matching for challenging environments. IEEE Transactions on Intelligent Transportation Systems , 18(4):847--857, 2016

    2016

  17. [25]

    Storing line segments in partition trees

    Mark H Overmars, Haijo Schipper, and Micha Sharir. Storing line segments in partition trees. BIT Numerical Mathematics , 30(3):385--403, 1990

    1990

  18. [26]

    Development of a weight-based topological map-matching algorithm and an integrity method for location-based ITS services

    Nagendra R Velaga. Development of a weight-based topological map-matching algorithm and an integrity method for location-based ITS services . PhD thesis, Nagendra R. Velaga, 2010

    2010

  19. [27]

    Gtfs bus stop mapping to the osm network

    Jan Vuurstaek, Glenn Cich, Luk Knapen, Wim Ectors, Tom Bellemans, Davy Janssens, et al. Gtfs bus stop mapping to the osm network. Future Generation Computer Systems , 110:393--406, 2020

    2020

  20. [28]

    An adaptive map-matching based on dynamic time warping for pedestrian positioning using network map

    Yuki Wakuda, Satoshi Asano, Noboru Koshizuka, and Ken Sakamura. An adaptive map-matching based on dynamic time warping for pedestrian positioning using network map. In Proceedings of the 2012 IEEE/ION Position, Location and Navigation Symposium , pages 590--597. IEEE, 2012

    2012

  21. [29]

    Map matching by fr \'e chet distance and global weight optimization

    Hong Wei, Yin Wang, George Forman, and Yanmin Zhu. Map matching by fr \'e chet distance and global weight optimization. Technical Paper, Departement of Computer Science and Engineering , 19, 2013

    2013

  22. [30]

    Some map matching algorithms for personal navigation assistants

    Christopher E White, David Bernstein, and Alain L Kornhauser. Some map matching algorithms for personal navigation assistants. Transportation research part c: emerging technologies , 8(1-6):91--108, 2000

    2000

  23. [31]

    Deepmm: Deep learning based map matching with data augmentation

    Kai Zhao, Jie Feng, Zhao Xu, Tong Xia, Lin Chen, Funing Sun, Diansheng Guo, Depeng Jin, and Yong Li. Deepmm: Deep learning based map matching with data augmentation. In Proceedings of the 27th ACM SIGSPATIAL international conference on advances in geographic information system...

    2019

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.