Graph-based metadata modeling in indoor positioning systems
Introduction
With the uprising of the Internet of Things (IoT), buildings (public, commercial, residential) were the main focus of innovative automated solutions. Resource usage efficiency attracts the most attention from both research and industry communities [24], resulting in the creation of smart Building Management Systems (BMSs) with functionalities steered towards optimizing how resources are utilized. While feeding different contextual metadata benefits the indoor positioning system’s (IPS) efficiency (increase overall accuracy, derive occupancy or movement patterns to create energy-saving strategies, etc.), the idea behind BMSs is also to address the well-being (social, comfort, etc.) of tenants.
IPSs are gaining the attention of the research community and industry that try to solve challenges in the domain of smart spaces. Data obtained from indoor positioning systems provide insightful information about the usage of the observed space. Other information, derived from positioning data, can be considered as metadata in IPS. Metadata further enhances positioning data by providing contextual insight, such as floor or room information, space occupancy, energy consumption, and similar.
It can be challenging to derive positioning/occupancy without applying some intrusive, high cost-inducing approaches like video surveillance, dedicated sensory systems, and wearable devices [29]. To achieve efficient real-time monitoring these approaches might be more acceptable, but come at the expense of high cost, privacy issues, and reduced scalability. IPS based on WiFi or Bluetooth are inexpensive and scale well. Additionally, WiFi-based IPS are easily deployed on existing infrastructures. Privacy of positioning information in such systems is preserved by of-loading location computation functions to most convenient devices (e.g. in edge computing IPS location information does not leave the local network). Through privacy intrusion may be minimized in such systems, it is not eliminated; hence, privacy remains a challenge [35]. Fortunately, modern mobile users are always carrying their wireless devices (smartphones, wearables), and through them are in continuous communication with the networking infrastructure inside and around a building. Thus, by observing the physical dispersion and continuous movements of such mobile devices, the behavioral characteristics of users of the indoor space can be inferred.
Characterizing social behavior inside a network of people is a challenging task. For public/company buildings it is more straightforward since there are established assumptions/patterns of typical behavior (e.g. daily routines of employees, restricted movement rules for visitors, meal breaks at certain times, etc.). Such assumptions cannot be made inside a residential building since tenants’ behavior is highly stochastic – movement patterns of tenants inside private residential buildings are rather inconsistent, without periodicity and ad-hoc, as opposed to public/company buildings where the allocation of employees and movement patterns are rather consistent. Therefore it is harder to model and extrapolate useful information from private residential buildings. Furthermore, observing the formation and evolution of social groups, which is a highly relevant problem in social networks, is significantly harder to maintain for such ad-hoc, stochastic networks modeling human social behavior.
By modeling social behavior inside residential buildings, IPS can enhance the underlying BMS with the information enabling them to animate and support meaningful interaction between proximate users, network serendipitous social encounters, and to seamlessly integrate events with the way interaction takes place in the observed social network. BMSs can also use this information to drive business decisions such as building a common entertainment room, organizing indoor social events, targeted community marketing (e.g. offer event tickets), etc.
All of the above-mentioned challenges in modeling social behavior for IPS are addressed with Bluetooth Low Energy Microlocation Asset Tracking system (BLEMAT). BLEMAT is a semi space-agnostic, a context-aware edge computing system that performs real-time indoor positioning, smoothing and filtering, fingerprinting, and floor plan layout detection supplemented with various complex data analytics and forecasting tasks. Additionally, BLEMAT is equipped with occupancy detection, neural network-based forecasting, and patterns extraction algorithms [26], [27].
In this article we are proposing enhancements to BLEMAT, a series of novel graph-based approaches for modeling social behavior data in indoor spaces. Our approach to modeling of tenants’ movement paths and detecting the existence of patterns is transforming graphs to sentences to apply less time-consuming similarity and pattern extraction algorithms. In our approach to modeling of tenants’ social relationships, we add several relevant vertices/edges attributes such as frequency, quality, and consistency of detected relationship. On social relationship graphs, we showcase two distinctive approaches to the detection of social communities and offer comparison metrics and discussion. For the time-continuous detection of social communities, we provide a novel approach to track their evolution (major events that lead to communities merging or dissolution) by vertex label propagation.
Finally, datasets, source code, and original high-resolution figures from data analysis performed in this article are published as open-source on Zenodo [36]. Datasets from this research are a contribution to quality real-world positioning datasets, and hopefully, they will be used by other researchers to test approaches from this article, as well as other approaches. As most of the publicly available positioning and occupancy datasets come from public places (universities, shopping malls, etc.), a quality dataset from a regular residential building is an important addition to the research field of IPS.
The rest of the article is structured as follows. Section 2 elaborates on related work in three IPS-related research areas: metadata modeling approaches, movement patterns exploration, and social networks and communities. Section 3 will underline our contributions to IPS metadata modeling through BLEMAT and present theoretical foundations for two types of positioning data graphs we use in the rest of the paper. In Section 4, we present the experimental environment (real-world BLEMAT deployment site) that is referenced throughout the article and the occupancy metadata collection workflow. Section 5 presents and elaborates on the practical usage of the graph described in Section 3 inside the BLEMAT IPS. Section 6 is the largest section, and it presents in detail the performed experiments on the observed deployment site by exploring tenant movement patterns and exploring tenant social relationships existence, quality, and grouping. Section 7 discusses the efficiency, novelty, and usability of the presented results. Finally, Section 8 contains conclusive and future-work remarks.
Section snippets
Related work
Information retrieval from IPS is a complex task. Positioning data can be obtained through a multitude of devices and sources: video surveillance [23], WiFi, Bluetooth [20] and/or sensors: lighting [31], pressure [39], audio [22], but also inertial sensors such as accelerometers and heading sensors [6]. Non-intrusive variants (i.e. observing signal behavior) are preferred since they better preserve the privacy of observed space, as well as deployment costs. For a cost-efficient IPS, it is
Graph-based metadata modeling in BLEMAT
BLEMAT is a Bluetooth Low Energy (BLE)-based IPS that rests on a deployed infrastructure of edge gateways called scanners. BLE beacons are the devices being tracked in the system. Scanners are capable of scanning the environment for active beacons and calculating their position in the space. Beacons are mobile and are emitting Bluetooth signal, also known as advertising.
In this article, we propose algorithmic improvements to BLEMAT, a collection of graph-based approaches for modeling social
Experimental setup
In our previous papers [26], [27] following innovative research achievements were presented: BLEMAT’s satisfiable accuracy and precision according to academic benchmarks, positive region-coverage density impact, positive impact in obstacle detection via RSSI signal deviation, and the positive effect of utilizing the Kalman filter for consecutive asset tracking and indoor positioning, and occupancy detection, extraction, and forecasting, using only wireless Bluetooth and internet links in a
BLEMAT graph-based location metadata modeling
From the occupancy dataset shown in Fig. 4 tenant path and behavior graphs can be generated (for a given period). For graphs management in BLEMAT networkx,1 igraph2 and graphviz3 libraries are used in Python. Visualization is an important aspect of every data science task. If the data is visualized accordingly, it is more understandable, and decision-making processes at higher levels (i.e. company management, not algorithmic) are of better quality.
Determining existence of tenant movement patterns using string similarity measures
In our previous paper, we have elaborated in detail how patterns of occupancy of apartments in a residential building can be extracted from raw 3D positioning data [27]. Occupancy patterns differ significantly from movement patterns – the former describe how the indoor space is occupied, and the latter how users of the indoor space interact with it. These movement patterns represent interaction patterns both between users and users and the indoor space. Detecting the existence of movement
Discussion
With performed experiments, we have showcased how different IPS metadata can be extracted and used to model the behavior of tenants in a residential building. In this section, we discuss the applicability of achieved experimental results, with emphasis on impact from 3 aspects of BLEMAT IPS: (1) Tenant movement paths, (2) Tenant behavior graphs, and (3) Computational efficacy.
(1) Tenant movement paths – By modeling tenant movement paths, we are providing a visual overview of tenant’s movement
Conclusion and future work
In this article, we have proposed algorithmic upgrades to BLEMAT, a set of graph-based approaches for modeling social behavior data: modeling of tenants’ movement paths and detecting the existence of patterns, modeling of tenants’ social relationships (frequency, quality) as well as detecting social communities and tracking their evolution. Since offering such services is not typical for an IPS, these enhancements make BLEMAT stand-out as a comprehensive and highly capable IPS, with a plethora
Acknowledgments
The authors acknowledge financial support of the Ministry of Education, Science, and Technological Development of the Republic of Serbia (Grant no. 620 451-03-68/2020-14/200125).
References (45)
On a relation between graph edit distance and maximum common subgraph
Pattern Recognit. Lett.
(1997)- et al.
Can social influence drive energy savings? Detecting the impact of social influence on the energy consumption behavior of networked users exposed to normative eco-feedback
Energy Build.
(2013) - et al.
Approximate graph edit distance computation by means of bipartite graph matching
Image Vis. Comput.
(2009) - et al.
Social networks
Encycl. Quant. Risk Anal. Assess.
(2008) - et al.
Local graph alignment and motif search in biological networks
Proc. Natl. Acad. Sci.
(2004) - et al.
Virtual social network communities: an investigation of language learners’ development of sociopragmatic awareness and multiliteracy skills
CALICO J.
(2011) - et al.
Fast unfolding of communities in large networks
J. Stat. Mech.
(2008) - et al.
Indoor positioning system using accelerometry and high accuracy heading sensors
Proceedings of the ION GPS/GNSS 2003 Conference
(2003) - et al.
Graph similarity features for HMM-based handwriting recognition in historical documents
Proceedings of the 12th International Conference on Frontiers in Handwriting Recognition
(2010) Facilitating social networking in inner-city neighborhoods
Computer
(2006)
Learning indoor movement habits for predictive control
Int. J. Space-Based Situated Comput.
Community structure in social and biological networks
Proc. Natl. Acad. Sci.
Graph Theory and its Applications
Graph model based indoor tracking
Proceedings of the Tenth International Conference on Mobile Data Management: Systems, Services and Middleware, MDM’09
Applied Graph Theory in Computer Vision and Pattern Recognition
Cross-layer design and optimization techniques in wireless multimedia sensor networks for smart cities
Comput. Sci. Inf. Syst.
Rfid-based indoor location tracking to ensure the safety of the elderly in smart home environments
Pers. Ubiquit. Comput.
Mobile sensing of pedestrian flocks in indoor environments using WiFi signals
Proceedings of the 2012 IEEE International Conference on Pervasive Computing and Communications
An application of Markov jump process model for activity-based indoor mobility prediction in wireless networks
Proceedings of the 2011 Frontiers of Information Technology
Algorithms for graph similarity and subgraph matching
Proceedings of the 2011 Ecological Inference Conference
Survey of wireless indoor positioning techniques and systems
IEEE Trans. Syst. Man Cybern. Part C (Appl. Rev.)
How much topological structure is preserved by graph embeddings?
Comput. Sci. Inf. Syst.
Cited by (2)
Modelling and visualizing people flow in smart buildings: A case study in a university campus
2021, GoodIT 2021 - Proceedings of the 2021 Conference on Information Technology for Social GoodLegal Issues and the Need to Engineer Positioning Systems for Protection of Privacy and Personal Security
2021, EAI/Springer Innovations in Communication and Computing