Map-Matching Queries under Fréchet Distance on Low-Density Spanners

TL;DR

This work addresses map-matching under the Fréchet distance by preprocessing a road-network graph to support fast queries of the form minπ d_F(π,Q) between any graph path π and a query polygonal curve Q, plus the ability to report the minimizing path. Departing from prior c-packed assumptions, the authors require the map to be a λ-low-density, t-spanner graph, a model closer to real-world networks, and they replace a semi-separated pair decomposition with a hierarchical separator framework. They present a (1+ε)-approximate data structure that achieves subquadratic space and near-linearithmic preprocessing, along with efficient reporting of the matching path; the core results hinge on using small balanced separators (τ-lanky graphs) and transit-vertex techniques. For fixed endpoints and segments, they obtain a 3-approximation and then extend to general polygonal queries, including edge-sampling constructions for alignment. Overall, the paper delivers near-optimal practical guarantees under realistic network assumptions and outlines open questions about parameter choices and potential lower bounds.

Abstract

Map matching is a common task when analysing GPS tracks, such as vehicle trajectories. The goal is to match a recorded noisy polygonal curve to a path on the map, usually represented as a geometric graph. The Fréchet distance is a commonly used metric for curves, making it a natural fit. The map-matching problem is well-studied, yet until recently no-one tackled the data structure question: preprocess a given graph so that one can query the minimum Fréchet distance between all graph paths and a polygonal curve. Recently, Gudmundsson, Seybold, and Wong [SODA 2023, arXiv:2211.02951] studied this problem for arbitrary query polygonal curves and $c$-packed graphs. In this paper, we instead require the graphs to be $λ$-low-density $t$-spanners, which is significantly more representative of real-world networks. We also show how to report a path that minimises the distance efficiently rather than only returning the minimal distance, which was stated as an open problem in their paper.

Figures (4)

• Figure 1: An example road network, in Barcelona. The total road length in a disk of radius $r$ is closer to $cr^2$ than $cr$, so this road network is not $c$-packed. However, the number of long edges intersecting the purple disk is small, and the red path is not much longer than the blue path, so the network is $\lambda$-low density and a $t$-spanner for small $\lambda$ and $t$. Map data from OpenStreetMap openstreetmap.
• Figure 2: A representation of the hierarchy (left) for the graph (right). A query segment is shown in purple, a possible path in blue. We check nodes $5$, $2$, and $1$. If we pick the transit vertex in $5$, then the path may be $10 \to 2 \to 5$, so we may need to go up the tree to find the next transit pair.
• Figure 3: A query trajectory $pq$ is shown in purple, and the reported path in the graph is shown in blue. We sample points on $pq$ at regular distance and snap them to the exponential grid around the graph vertices. Once we find $r$ on $pq$ that aligns with the transit vertex $s$, we can query the pair $(u, s)$ with the (snapped) segment $pr$ to find the next vertex $u'$.
• Figure 4: The surface of a trough: at fixed $z$, we include all points no further than $4z$ from $e$.

Theorems & Definitions (19)

• Definition 1
• Theorem 3
• Definition 4
• Definition 5
• Lemma 7
• Lemma 8
• Theorem 9
• Lemma 11
• Lemma 12
• Lemma 13