Continuous Map Matching to Paths under Travel Time Constraints

TL;DR

This paper introduces a novel algorithm for map matching with travel time constraints that handles infinite candidate locations, guarantees solution detection, and outperforms classical methods in efficiency.

Contribution

We propose a new algorithm for map matching under travel time constraints that can handle infinite candidate sets and guarantees to find a consistent path if one exists.

Findings

Algorithm always detects a consistent map matching path if it exists.

Our algorithm has better theoretical and practical running time than baseline methods.

Experimental results demonstrate the algorithm's effectiveness on real and synthetic data.

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.