The construction and update of a radio map are usually referred as the main drawbacks of WiFi fingerprinting, a very popular method in indoor localization research. For radio map update, some studies suggest taking new measurements at some random locations, usually from the ones used in the radio map construction. In this paper, we argue that the locations should not be random, and propose how to determine them. Given the set locations where the measurements used for the initial radio map construction were taken, a subset of locations for the update measurements is chosen through optimization so that the remaining locations found in the initial measurements are best approximated through regression. The regression method is Support Vector Regression (SVR) and the optimization is achieved using a genetic algorithm approach. We tested our approach using a database of WiFi measurements collected at a relatively dense set of locations during ten months in a university library setting. The experiments results show that, if no dramatic event occurs (e.g., relevant WiFi networks are changed), our approach outperforms other strategies for determining the collection locations for periodic updates. We also present a clear guide on how to conduct the radio map updates.
The performance of indoor positioning systems is usually measured by their accuracy in meters. This facilitates the comparison of different systems, but does not necessarily give information about how well they perform in real-life scenarios, e. g. during indoor navigation of walking persons. In this paper, we present a task-oriented evaluation that adapts the idea of landmark navigation: Instead of specifying the error metrically, system performance is measured by the ability to determine the correct segment of an indoor route, which in turn enables the navigation system to give correct instructions. We introduce the area match metric in order to identify areas where positioning proves problematic. In order to evaluate the described metric, we use a pedestrian dead reckoning approach to compute indoor positions. Without any external correction, the correct segment of the test route is identified in 88.4% of all trials. Based on these results, we explore options how to identify and predict erroneous situations during the navigation process as well as beforehand.
This research combines spatiotemporal traffic and population distribution data in a dynamic isochrone map. To analyze the number of people who have access to a given area or location within a given time, two spatiotemporal variations should ideally be taken into account: (1) variation in travel time, which tend to differ throughout the day as a result of changing traffic conditions, and (2) variation in the location of people, as a result of travel. Typically, accessibility research includes neither one or only variation in travel time. Until recently, we lacked insight in where people were located throughout the day. However, as a result of new data sources like GSM data, the opportunity arises to investigate how variation in traffic conditions and variation in people's location influences accessibility through space and time. The novelty of this research lies in the combination of spatiotemporal traffic data and spatiotemporal population distribution data presented in a dynamic isochrone web map. A case study is used for the development of this isochrone map. Users can dynamically analyze the areas and people who can reach various home interior stores in the Netherlands within a given time, taking into account traffic conditions and the location of people throughout the day.
A fundamental problem in indoor location-based services is to compute the meaning of location with respect to an indoor location model. One specific challenge in this area is represented by the central tradeoff between two philosophies: a decent amount of the community tries to provide high-quality, high-fidelity models investing specialized knowledge and a lot of time in building such models for each building thereby increasing simplicity and quality of location-based services such as navigation or guidance. In contrast to that, other people argue that crowd sourcing and very simple representations of environmental information are the only way of generating indoor environmental information at scale. However, applications then have to tolerate errors and deal with oversimplified models. With this paper, we show for a specific widely accepted simple environmental model in which building floorplans are represented as black-and-white bitmaps, how we can provide algorithms for extracting higher order topological concepts from these trivial maps. We further illustrate how these can be applied to the hard problem of indoor shortest path calculation, indoor alternative path calculation, indoor spatial statistics, and path segmentation.
Autonomous mobile robots show promising opportunities as concrete use cases of location-based services. Such robots are able to perform various tasks in buildings using a wide array of sensors to perceive their surroundings. A connected area of research which forms the basis for a deeper understanding of these perceptions is the numerical representation of visual perception of space. Different structures in buildings like rooms, hallways and doorways form different, corresponding patterns in these representations. Thanks to recent advances in the field of deep learning with neural networks, it now seems possible to explore the idea of automatically learning these recurring structures using machine learning techniques. Combining these topics will enable the creation of new and better location-based services which have a deep awareness of their surroundings. This paper presents a framework to create a data set containing 2D isovist measures calculated along geospatial trajectories that traverse a 3D simulation environment. Furthermore, we show that these isovist measures do reflect the recurring structures found in buildings and the recurring patterns are encoded in a way that unsupervised machine learning is able to identify meaningful structures like rooms, hallways and doorways. These labeled data sets can further be used for neural network based supervised learning. The models generated this way do generalize and are able to identify structures in different environments.
