Correction to ‘Open source map matching with Markov decision processes: A new method and a detailed benchmark with existing approaches’

Wöltche, A. (2023). Open source map matching with Markov decision processes: A new method and a detailed benchmark with existing approaches. Transactions in GIS, 27, 1959–1991. https://doi.org/10.1111/tgis.13107

The numbering of the list on pages 1960 to 1961 in the Introduction is incorrect. The correct numbering of this list is stated below.

• 1.1 Explains the terms we use in the remainder of this work.
• 1.2 Illustrates the main challenges of map matching.
• 1.3 Provides an overview of relevant literature and state-of-the-art Open Source Software (OSS).
• 1.4 Introduces our novel approach and displays the challenges we specifically address.
• 1.5 Gives a real-world example of how the current state-of-the-art compares to our new approach, which we describe as follows.
• 2 Gives the technology roadmap and explains our new approach, that is, a combination of several new and improved technologies:
• 2.1 Introduces our custom Trajectory Simplification (TS) algorithm that is used for FCD preprocessing. We combine a customized Douglas-Peucker algorithm with a custom point-cluster-merging algorithm to reduce simple noise patterns before actual map matching to improve performance.
• 2.2 Explains Candidate Search (CS), which is used to select possible mapping locations in the given road network for a given track. CS selects a candidate set for each point in a track by pointing to spatially nearby road locations (candidates). From each pair of adjacent candidate sets, one candidate is chosen so that a route between them (candidate route) can be computed.
• 2.3 Introduces our novel Candidate Adoption (CA) feature that depends on CS. It takes into account for each candidate set additional candidates from the surrounding candidate sets. This allows our map-matching algorithm to stochastically handle even large outliers and high noise of tracks in order to further improve accuracy.
• 2.4 Introduces our new comparison algorithm that allows to extract and evaluate the differences and similarities of alternative routes within the same road network, for example, matches to a given ground truth. With its ability to handle small inaccuracies between the given data, this algorithm is used in our benchmark for comparing the results of our new approach with existing solutions.
• 2.5 Introduces our new map-matching model based on Markov decision processes (MDPs) and Reinforcement Learning (RL) algorithms. The MDP uses absolute rewards for optimizing map-matching solutions. These rewards are calculated with our new map-matching metrics which evaluate direction changes on the candidate routes in addition to distances and lengths between track segments and candidate routes. Direction changes facilitate CA to penalize detours and thus lead to collapsing outliers and high noise clusters.
• 3 Evaluates the performance of our novel map-matching approach in multiple experiments on several data sets. We compare our approach against six existing third-party OSS solutions and discuss the results of this benchmark.
• 4 Finally provides an outlook on future research our novel approach enables.

The typesetter regrets the error.