IEEE Geoscience and Remote Sensing Magazine - March 2021 - 194

Between January and May 2019, the number of active
OSM contributors per month was fewer than 1% of registered users. However, several efforts have been made to
encourage volunteers to frequently edit data. Gamification
strategies could be applied to solve the problem by assigning
annotation tasks to volunteers with a gamelike interface and
scoring systems. Some gamification projects have been listed in the OSM wiki [116], but they are not in widespread use.
Contributor analysis in OSM has been studied in several
works [97]. It has been observed that volunteers' experience
and familiarity with the edited area of the map are good
proxies to estimate the quality of their annotations in OSM.
The authors of [98] propose a measure for estimating annotation trust using annotation statistics obtained from volunteers' activities, object geometries, and temporal data. A
trust index could be also obtained through machine learning methods using OSM data statistics and some reference
data. Such a trust index could help improve methods created to verify the correctness of labels of objects registered
in OSM [54], [56].
NEW WAYS OF SEARCHING AND INTERACTING
WITH OSM
Search tools are important features of mapping services.
However, the search tool provided by the OSM website has
limited capabilities-essentially, finding the location of a
given place name or address. In [99], the authors propose
NL Maps [117], a natural-language interface to query data
in OSM. This service can answer textual questions about
geographical facts in OSM. The response includes a text
message and a map that highlights geographical objects of
interest. For instance, NL Maps can answer the following
question: What is the closest supermarket from the Royal
Bank of Scotland in Edinburgh? Figure 8 illustrates the
search results in NL Maps for the question, What museums
are there in Heidelberg? First, the names of the found museums are shown and then their locations are presented with
markers on the OSM map of the city of Heidelberg.

FIGURE 8. NL Maps search results for the query, What museums

are there in Heidelberg? (Source: NL maps.)
194

This work uses a semantic parser for the OSM data, which
was proposed in [100] and extended in [101], to transform
natural-language text to a machine-readable language formula [102]. This representation is used to create a structured query and retrieve OSM information employing an
extension of the Overpass application programming interface (API): Overpass-NL Maps [118]. Although NL Maps
can handle several common questions, it is not able to answer complex questions such as, Where are the four-star
hotels in Berlin? The authors of [101] also propose to improve the NL Maps performance through user feedback. A
simple example of feedback is to ask the user if the result of
the query is helpful or not. More complex feedback can be
obtained from expert users who have knowledge of OSM
and the Overpass API. For instance, one can ask the expert user if the intermediate results of the query-processing
pipeline are correct or not. We believe that the implementation of a more intelligent tool for answering natural-language queries on the OSM website could potentially attract
more users.
Recently, the authors of [103] used OSM data for the
Visual Question Answering (VQA) problem, a task that
consists of providing a natural-language answer for a given
image and a natural-language question about the image.
The authors create a data set using OSM data and Sentinel-2
satellite imagery for training a VQA model. The VQA is a
challenging problem, and its application to remote sensing
imagery is still in its infancy. It can enable some queries in
places where the mapping data are incomplete but aerial
imagery is available.
OSM-ENABLED AUGMENTED REALITY
Recent applications of OSM data can be further improved
using machine learning methods. For instance, augmented reality (AR) has great potential to improve the way we
experience cities, especially for tourism. In [104], an AR
tool is presented to recommend city POIs by taking into
account user profiles. The authors propose a mobile application that shows POIs registered in OSM in real time,
with the mobile camera view as background. A case study
in the city of Trani, Italy, shows how a tourist interested
in local architectural work can visualize the POIs in the
mobile application (within a distance radius) marked as
colored circles and with labels superimposed across the
mobile camera view. In [104], a hand-engineered rule is
proposed to define which POIs to show the user. For this
type of AR application, reference data could be employed
by asking for users' feedback (e.g., asking a user to add
POIs that he or she has visited to a favorites list). Thus,
obtaining reference data can enable the use of more effective supervised-learning methods to predict adequate
POIs for new users.
CONCLUSIONS
In this article, we reviewed synergetic efforts involving OSM
and machine learning. In the first part, we summarized
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

MARCH 2021

IEEE Geoscience and Remote Sensing Magazine - March 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2021