Employing Spatial Analysis in Indoor Positioning and Tracking Using Wi-Fi Access Points

Song Gao Sathya Prasad Department of Geography Applications Prototype Lab University of California, Santa Barbara Esri Inc., Redlands CA 93106, USA CA 92373, USA [email protected] [email protected]

ABSTRACT Bluetooth sensors [29, 33]. Positioning is the key compo- A practical WiFi-based has to be adapt- nent to support smart mobility studies, trajectory data min- able to the variations of indoor environmental dynamic fac- ing and travel related applications [43, 36, 40]. It is well tors. In this work, we propose a novel Wi-Fi indoor posi- known that GPS-enabled devices work well for positioning tioning and tracking framework which employs the spatial in outdoor environments but not in indoor environments be- analysis and image processing techniques. The Wi-Fi sur- cause of the signal obstructions and attenuation[26]. How- faces can be dynamically constructed and updated and thus ever, as reported in a national human activity pattern survey help to address the challenges of signal spatial heterogene- (NHAPS), people spend an average of 87% of their time in ity and environmental variations. A mobile app for indoor enclosed buildings and about 6% of their time in enclosed positioning application has been developed as a proof of con- vehicles [19]. There is a high-demand for indoor positioning cept. Based on the experiments we conducted at the Esri technologies. In [44] the researcher systematically compare campus, this method can achieve about 2-meter positioning three dominant positioning technologies: Assisted-GPS, Wi- accuracy. The proposed methodology and theoretical frame- Fi, and Cellular positioning. Their pros and cons are dis- work can guide engineers to implement cost-effective indoor cussed in terms of coverage, accuracy and reliability. It re- positioning infrastructure, and thus offer insights on future ports that Assisted-GPS obtains an average median error of smart campus applications. The introduced spatial analy- 8m outdoors while Wi-Fi positioning only gets 74m of that sis and geoprocessing workflow may also bring the attention and cellular positioning has about 600m median error in av- of GIScientists to make more efforts to conquer the indoor erage and is least accurate. However, high-resolution GPS or positioning and tracking challenges. Assisted-GPS positioning chipsets don’t work well in indoor environments due to limited satellite visibility; and thus a variety of indoor positioning technologies and systems have CCS Concepts been designed and developed to increase the indoor posi- •Information systems → Mobile information pro- tioning accuracy. cessing systems; Spatial-temporal systems; Loca- The widely use of Wi-Fi access points for Internet connec- tion based services; Geographic information sys- tion in hotels, offices, coffee shops, airports and many other tems; Global positioning systems; fixed places makes Wi-Fi become an attractive technology for the positioning purpose. Location positioning systems Keywords using area local network (WLAN) infrastructure are considered as cost-effective and practical solutions for Indoor positioning; Spatial analysis; WiFi fingerprinting indoor location estimation and tracking [7]. There is almost no extra hardware or other infrastructure investment, which 1. INTRODUCTION is different from other indoor positioning technologies such In the era of mobile age, location-based systems and ser- as low-energy Bluetooth sensor networks or radio-frequency vices become more and more popular in people’s daily life, identification (RFID) systems. Conventional methods for such as searching nearby points of interest (POI), way find- indoor positioning are based on time of arrival, time differ- ing and . There has been growing interest among ence of arrival and angle of arrival of radio signals transmit- researchers in studying human mobility patterns based on ted by mobile stations [30]. Another type of WLAN posi- the data collected from location-awareness devices, e.g., tioning method unitizes the fingerprinting idea to determine GPS-enabled devices [45, 39], cellular phones [13, 18], and a receiver’s location via clustering and probabilistic distri- butions [42]. The accuracy of localization is dependent on Permission to make digital or hard copies of all or part of this work for personal or the separation distance between two-adjacent Wi-Fi refer- classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation ence points and the transmission range of these reference on the first page. Copyrights for components of this work owned by others than the points [5]. One challenge that WLAN indoor positioning author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or techniques face is the spatial heterogeneity of Wi-Fi signal republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from [email protected]. surfaces caused by the signal path loss because of absorp- ISA 16, October 31-November 03 2016, Burlingame, CA, USA tion, reflection and refraction by the surrounding environ- c 2016 Copyright held by the owner/author(s). Publication rights licensed to ACM. ment. Thus the physical radiation models need complex ISBN 978-1-4503-4585-9/16/10. . . $15.00 parameter estimation and empirical fitting. Spatial analysis DOI: http://dx.doi.org/10.1145/3005422.3005425 techniques can be employed in dealing with geometric, topo- In an indoor environment equipped with equal to or large logical, and geographic properties which can help to capture than three Bluetooth low energy (BLE) beacons, the loca- better indoor environment and to construct more accurate tion of a target mobile device with Bluetooth can be de- Wi-Fi surfaces for indoor positioning and tracking purpose. termined using classic positioning approaches (see Section In this research, we propose a novel geoprocessing-based 2.2 in detail). In this way, location-dependent triggers, no- framework and develop a mobile application for real-time tifications and tracking activities can be enabled by em- indoor positioning and tracking using Wi-Fi access points. ploying multiple BLE beacons. There are several popular The goal of this study is to explore how accurate Wi-Fi in- BLE beacon-positioning protocols and technology available door positioning with the support of spatial analysis can online, including Apple iBeacon1, Google Eddystone2 and achieve. The following of this paper is organized as follows. Gualcomm Gimbal3, which guide developers to implement In Section 2, we review some related work and discuss the up-to-date indoor positioning and tracking applications. state of the art in indoor positioning technologies. Our pro- Radio-frequency Identification (RFID): is a general posed theoretical framework towards indoor positioning and term used for a system that communicates using radio waves tracking, and detailed methods are introduced in Section 3. between a reader and an electronic tag attached to an ob- The system implementation and experiments are described ject. Comparing with Bluetooth technology, RFID systems in Section 4, which is followed by discussions in Section 5. usually comprise of readers and tags that store relatively We conclude and give a vision for future work in Section 6. limited information about the object such as location and attribute information. Those tags can be activated and send 2. LITERATURE REVIEW out stored information if they receive the signal from RFID readers within certain distance thresholds, which can be 2.1 Indoor Positioning Technologies used to estimate the reader’s location and to show relevant information. Currently RFID positioning and tracking sys- In general, indoor positioning technologies can be classi- tems are widely used for asset tracking, shipments tracking fied into two broad categories: radio-frequency-based (RF) in supply chains, and object positioning in retailing places and non-radio-frequency-based (NRF) technologies. The RF and shopping malls. group includes but not limited to WLAN, Bluetooth, and Because of sensor diversity and positioning challenges in RFID systems, while the NRF group contains ultrasound, various indoor environments, there is also an increasing magnetic fields, and vision-based systems. Researchers have trend towards combining and integrating different sensor made great efforts in the field of indoor positioning using networks to get a better spatial coverage and position ac- these sensors and technologies [17, 11, 23, 15, 21, 24, 20]. curacy than using single data source. In [9] researchers in- The spatial coverage area and positioning accuracy of those tegrate Wi-Fi and inertial navigation systems to get perfor- different technologies have been reviewed by [26]. Here, we mance close to meter by fusing pedestrian dead reckoning only briefly discuss the RF technologies that are most popu- and Wi-Fi signal strength measurements. lar in the current market share and face several challenging issues to investigate. 2.2 Indoor Positioning Methods Wi-Fi: The Institute of Electrical and Electronics Engi- neers (IEEE) 802.11 work group [14] documents the stan- The following methods usually work for both outdoor and dard use of Wi-Fi technology to enable wireless network indoor environments, but we emphasize the indoor case here. connections in five distinct frequency ranges: 2.4 GHz, 3.6 Note that those methods can be applied in most radio- GHz, 4.9 GHz, 5 GHz, and 5.9 GHz bands. The wireless frequency-based sensors, such as Wi-Fi and Bluetooth. networks are widely implemented in many types of indoor 2.2.1 Geometric Approaches buildings in which wireless access points are usually fixed at certain positions. Those access points allow wireless de- In geometry, trilateration is a method to determine the vices (e.g., mobile phones, laptops and tablets) to connect target location of a point by measurement of distances to to a wired network using Wi-Fi technology. And the relative three points at known locations using the geometry of cir- distance between wireless devices and access points can be cles, triangles, or spheres [6]. This method has been widely roughly estimated based on Wi-Fi signal strength using sig- used not only in positioning systems [16], but also in robot nal propagation models [28]. It is also well known that the localization, computer graphics, aeronautics, and so on [34]. accuracy of indoor position estimation based on Wi-Fi signal For the positioning purpose, if already knowing the coor- strength is affected by many environmental and behaviour dinate (lat,lon) information of three fixed access points, it factors, such as walls, doors, settings of access points, orien- needs to convert the latitude and longitude of these loca- tation of human body, etc.[11, 37]. In practical applications, tions from the Earth reference system to axis values (x,y,z) a good approximation of heterogeneous environmental sig- in the Cartesian coordinate system as follows: nal surfaces could help to improve the indoor positioning x = R ∗ cos(lat) ∗ cos(lon) accuracy. Spatial regression which is a widely used spatial- y = R ∗ cos(lat) ∗ sin(lon) (1) analysis method in finding spatial patterns and modeling trend surface [8] could potentially be a good candidate. z = R ∗ sin(lat) Bluetooth: is designed for low power consumption and Where R is the approximate radius of earth (e.g., allows multiple electronic devices to communicate with each 6371km). Then we can try to find the solutions to trilat- other without cables by using the same 2.4 GHz radio- eration equations to approximate the location of the target frequency band as Wi-Fi. The distance range within which Bluetooth positioning can work is about 10 meters. In the 1https://developer.apple.com/ibeacon/ beaconing mode, Bluetooth permitted messages can be used 2https://developers.google.com/beacons/ to detect the physical proximity between two devices [10]. 3http://www.gimbal.com/ points. The set of access points and their signal strengths are distinctive and present a “fingerprint” to a unique loca- tion on the “radiomap”. Thus, we can employ searching and comparison processes in the online positioning phase to find the most probable location. Several fingerprinting matching and calculation methods have been developed to estimate the location of a user based on received Wi-Fi signal strength (SS) values at known ref- erence point (RP) locations [17, 22, 27, 12]. In its simplest form, it can be expressed mathematically as follows:

v u m uX 2 argminp∈(1,2,n)t [SSRE (i, p) − SSME (i)] (2) i=1

where SSRE (i, p) represents the received signal-strength value of access point i at a known reference point location p on the radiomap, and SSME (i) is the measured signal- strength value of access point i at the current unknown lo- cation. The location p that has the minimum root-squared- Figure 1: Determine the location of an object using differences of SS values for all available m access points be- trilateration method. tween reference points and the target location is considered as the most probable estimated location. object by referring to three points with known locations [16] In addition, machine-learning-based fingerprinting tech- (See Figure 1). The equations are nonlinear and it is not niques have also been studied for improving the quality of so easy to obtain an exact solution. Several iterative arith- location estimation in complex real environment, including metic methods have been proposed to find efficient solutions Bayesian modeling, k-nearest-neighbor estimation, support for trilateration-based localization [25, 41]. vector machine, neural networks and so on [2, 31, 42, 4]. It is obvious that the distance between a wireless device and an access point is the key for Wi-Fi positioning using the 3. METHODOLOGIES trilateration method. The received signal strength doesn’t directly lead to an estimated distance to an access point. In 3.1 Study Design general, it does follow a trend that the signal strength de- creases with an increase of distance as is expected, but it is Our theoretical framework of indoor positioning and not a simple linear path loss model. The log-distance path tracking consists of three steps as shown in Figure 2. loss model is one of the most simplistic radio-propagation Step 1: Field sampling is a process of collecting available models [28, 32] that predict the received signal strength at indoor Wi-Fi access point set {AP1, AP2, APm} and the a certain distance inside a building or densely populated ar- corresponding received signal strength indicator (RSSI) eas. However, in many practical situations, many factors un- at various locations {Loc1, Loc2, ..., Locn} with different derlying radio propagation can contribute to the reflection, distances to each access points. The output of this step is a refraction, absorption and scattering of signals. It is diffi- two-dimensional matrix {RSSI(APi, Locj )} which can be cult to predict the received signal strength with a simplistic represented as: log-distance pass-loss model. Rather than creating a new  RSSI RSSI ... RSSI  pass-loss model, some researchers try to re-parameterize the  (AP1,Loc1) (AP1,Loc2) (AP1,Locn)   RSSI RSSI ... RSSI  classic log-distance model. For example, researchers found (AP2,Loc1) (AP2,Loc2) (AP2,Locn) ...... that at closer ranges (e.g., smaller than 5 meters) the expo-    RSSI RSSI ... RSSI  nent factor of propagation model would take a higher value, (APm,Loc1) (APm,Loc2) (APm,Locn) and thus a dual-distance model has been proposed to adjust Note that the matrix allows for ’null’ value element if the propagation model at larger distances for achieving a there is no signal for a given Wi-Fi access point at a certain better positioning accuracy [3]. location.

2.2.2 Fingerprinting Approaches Step 2: The fixed location of each Wi-Fi access point Although most Wi-Fi access points are located at the fixed Loc(APi) can be estimated based on a sample of high qual- locations of buildings, it is very hard to get their accurate co- ity signals (i.e., RSSI value ≥ -50 decibels). The geomet- ordinates and detailed digital floor maps because of privacy ric centroid of these clustered high quality signal locations concerns. Fingerprinting approach is a popular wireless- is stored as Loc(APi). Meanwhile, as shown in the geo- network positioning technology in metropolitan areas since processing workflow (Figure 3), the signal surface of each it doesn’t need the exact position information of Wi-Fi ac- Wi-Fi access point Surf(APi) can be built via spatial inter- cess points. Instead, it needs to construct an offline database polation (e.g., inverse distance weighted (IDW) method or that contains the signal strength distribution of Wi-Fi ac- Kriging techniques) of collected sampling points from Step cess points at known locations, i.e., the “radiomap”. For a 1. The AP locations and raster surfaces of these Wi-Fi ac- given location, it receives varying signal values from equal to cess points generated from this geoprocessing step are stored or less than the maximum number of available Wi-Fi access on the GIS Cloud server where data can be published as Figure 2: A geoprocessing-based framework of indoor positioning and tracking.

Figure 3: A geoprocessing workflow for generating Wi-Fi signal surface from samples. Yellow squares represent spatial analysis and data conversion tools while green ellipse shapes represent intermediate data layers and outputs. standard geospatial Web services. Figure 4 shows a Wi-Fi surface with 1m*1m grids generated by the spatial interpo- lation process. Note that a grid location in this surface is associated with an array of signal strength from a list of available Wi-Fi access points. Step 3: The developed mobile app can estimate a user’s indoor location based on a received Wi-Fi signal array [RSSI(APi)]. More detailed algorithms will be discussed in the following section. Moreover, we integrate the Fire- Figure 4: A Wi-Fi surface with 1m*1m grids gener- base4 real-time database for storing and synchronizing cross- ated by the spatial interpolation process. platform app’s indoor positioning data for enabling indoor mobility tracking. 3.2 Positioning Algorithms Several positioning algorithms have been developed for

4https://www.firebase.com/ Figure 5: The proposed heuristics-based indoor po- sitioning algorithm.

Wi-Fi positioning [42, 17, 44, 7] and can be categorized as: geometric techniques, statistical methods, fingerprinting and machine learning algorithms. We introduce another algo- rithm based on image filtering technique and then integrate it with two widely used algorithms: Trilateration and k- Figure 6: Filtered Wi-Fi surface pixels that meet nearest-neighbor in signal space (K-NNSS) [2] into our in- signal strength requirements are highlighted with door positioning system relying on a Wi-Fi surface which cyan color. keeps at least three image pixels that meets the multi-value filtering requirements. As shown in Figure 5, we propose a heuristics-based in- indoor location. door positioning algorithm in the online mode. It contains K-nearest-neighbors is a popular method used for classifi- three main phases. First, a device scans the cation and regression [1], in which an object is grouped to surrounding environment and gets available Wi-Fi access the most class by a majority vote of its k-nearest neighbors. points with their RSSI value array: [RSSI(APs)]. Second, In the K-NNSS algorithm, it first calculates the distances a multidimensional-value filtering technique is employed to between the current observed vector of RSSI values and all extract Wi-Fi surface pixels on which the RSSI values on RSSI vectors in the signal space. the surface across all available access points are within the signal strength ranges, which can be expressed as follows: v u m uX 2 distj (SSr,SSo) = t [SSr(i, j) − SSo(i)] (4) Loc ∈ ∃Loc(j){RSSI(APs)i − δ ≤ RSSI(Loc ,AP ) j i (3) i=1 ≤ RSSI(APs)i + δ, |∀i ∈ (1, 2, ..., N)} th where SSr(i, j) represents the RSSI of the i Wi-Fi access It assumes that there exist one or several pixel locations point at a known location j on the Wi-Fi signal surface, and [Locj ] on the surface such that its corresponding received SSo(i) is the measured RSSI of the same Wi-Fi access point signal strength is within a constant δ deviation away from i at the current unknown location. the current RSSI for each available Wi-Fi access point i (see After calculating all the signal-strength distances with re- Figure 6). Third, depending on the resulting number N of spect to all locations on the signal surface, a set of k loca- pixels left after the image filtering step, the program deter- tions with smallest distances is chosen. The chosen value of k mines which positioning approach should be further trigged. varies in different realistic environments because of changes If N = 0, the device needs to re-scan environmental Wi-FI in available Wi-Fi access points, and a calibration process signals until get updated RSSI values and then repeat pre- can help to derive an appropriate value. We choose k=5 vious processing steps; if 0 < N < 3, the algorithm will in empirical studies. Then the current unknown location simply return the mean coordinate of the remained surface (lato, lono) can be approximated by averaging the coordi- pixels as the estimated location; if N = 3, the geometric nate information of each candidate location (latj , lonj ) from method that utilizes the aforementioned propagation model those chosen k-nearest-neighbor locations in the signal space and trilateration positioning approach (see Section 2.2 for as follows: more details about this method) is chosen, and the loca- tions of three Wi-Fi access points with largest RSSIs are k 1 X taken as the reference points for trilateration; if N>3, the (lat , lon ) = ∗ (lat , lon ) (5) o o k j j k-nearest-neighbors algorithm is chosen to approximate the j=1 or by taking the spatial weights (distance) into considera- distance-based trilateration model in real-world complex tion, which is the chosen algorithm in developing our indoor- environments. But fingerprinting methods need database positioning mobile application: searching and comparison computation processes in the on- line mode, which takes long time if it involves large-size mul- k tidimensional signal surfaces. When choosing the trilatera- P 1 ∗ (latj , lonj ) distj (SSr ,SSo) tion method, the parameters in the propagation model for j=1 (lato, lono) = (6) converting signal strength to distance values need to be well k P calibrated. Otherwise sometimes no good matching results distj (SSr,SSo) j=1 can be returned. Last but not least, by applying the spatial analysis and where dist (SS ,SS ) is the previously calculated dis- j r o interpolation methods, it helps to approximate the spatial tance between two RSSI vectors. This algorithm is actually distribution of signal strength in real-world indoor environ- consistent with the inverse-distance weighted interpolation ments. But it also needs to try different spatial interpolation in spatial analysis, which estimates the grid-cell attribute methods and parameter fitting processes for choosing the values in a raster from a set of sample points by weighting best one for generating good signal surface representations. sample-point values so that the farther a sampling point is away from the target cell being evaluated, the less weight it contributes to the calculation of the target cell’s value. 6. CONCLUSIONS AND FUTURE WORK The method is inspired by Tobler’s first law of geography: A practical WLAN-based positioning system has to be “Everything is related to everything else, but near things are adaptable to the variations of indoor environmental dynamic more related than distant things.” [35]. factors. In this work, we propose a novel Wi-Fi indoor po- sitioning and tracking framework which employs the spatial 4. IMPLEMENTATION AND RESULT analysis and image processing techniques. The Wi-Fi sur- faces can be dynamically constructed and updated, and thus Based on the geopeocessing framework and workflow in- help to address the challenges of signal spatial heterogeneity troduced above, we first developed a Wi-Fi scanning mobile and environmental variations. In addition, we don’t need to application to scan and collect the available Wi-Fi access collect a very large pool of reference points for RSSI sam- points inside a building of Esri campus (Figure 7a). In the pling comparing with other signal-strength database con- experiments, we selected 18 Wi-Fi access points in a build- struction methods in the offline mode, although it is still ing and collected at least 30 sampling RSSI values for each necessary to get some Wi-Fi signal-strength samples from Wi-Fi access point (Figure 7b). Then the geoprocessing sys- indoor reference points. Instead, the weighted spatial in- tem was built for constructing the signal raster surfaces with terpolation method can help to reduce the field sampling 1-meter grid spatial resolution for all available Wi-Fi access burden. A mobile indoor positioning application has been points. In the online phase, the indoor position of the user developed as a proof of concept. Based on the experiments can be estimated via raster multi-value-attribute filtering we conducted at Esri campus, this method can achieve about on Wi-Fi signal surfaces and above introduced heuristics- 2-meter positioning accuracy. Since it falls into the general based positioning algorithm. As shown in Figure 7c, when category of “fingerprinting”, one can argue that the closer one mobile device was put at the location Q, the red circle we can approximate to the actual environmental signals, represented the estimated position based on our proposed the more accurate we can locate the user. The proposed method while the blue circle was the GPS location which methodology and theoretical framework can guide engineers still located on the outside of this building. After conduct- to implement cost-effective indoor positioning infrastructure ing several rounds of experiments, we got about 2-meter using Wi-Fi in other campuses, and thus offer insights on positioning accuracy in average in this empirical study. The future smart campus applications. The spatial analysis and user’ locations were also automatically sent to a real-time geoprocessing workflow may also bring the attention of GI- database (i.e. Firebase) for indoor mobility tracking. Scientists to make more efforts to conquer the indoor posi- tioning challenge. 5. DISCUSSIONS In future work, we aim to take the signal directionality Indoor positioning is a challenging problem and requires into consideration and try more advanced spatial interpo- a lot of interdisciplinary research. The biggest advantage lation and clustering methods to achieve better accuracy of Wi-Fi positioning technology is that it can utilize exist- from meter to decimeter. In addition, a crowdsourcing Wi- ing wireless-network infrastructures without extra hardware Fi signal collection and sharing platform will be developed to costs as Bluetooth beacons or RFID tags do need. How- engage more users’ contribution for advancing indoor posi- ever, it also has some limitations when applying in complex tioning technology in our campus. Last but not least, there indoor environments: is a growing trend towards integrating multi-sensors that First, the RSSI values vary dynamically and are sensitive are available in the mobiles phones, including speaker, ac- to indoor environments because of signal multi-path fading, celerometer, gyrometer, barometer etc., to improve indoor path loss, and even the direction to which human bodies are positioning accuracy and to semantically label indoor room facing. It needs a lot of work in the offline sampling process types. It is also fit into our future investigation into indoor to ensure that we can get a good approximation of Wi-Fi positioning and ambient sensing technologies. signal strength database at difference reference locations and surfaces with respect to the spatial structure. 7. REFERENCES Second, by comparing different positioning algorithms, [1] N. ALTMAN. An introduction to kernel and fingerprinting methods based on machine learning tech- nearest-neighbor nonparametric regression. The niques usually get a better position estimation than American Statistician, 46(3):175, 1992. Figure 7: Screenshots of the indoor positioning app using Wi-Fi scanning and spatial analysis techniques. (a) Wi-Fi scanning results at a given location; (b) available Wi-Fi access points in the building (green markers); (c) online-mode indoor positioning (red dot).

[2] P. Bahl and V. N. Padmanabhan. : An applications. In Proceedings of the 27th International in-building rf-based user location and tracking system. Technical Meeting of The Satellite Division of the In INFOCOM 2000. Nineteenth Annual Joint Institute of Navigation (ION GNSS+ 2014), Tampa, Conference of the IEEE Computer and FL, USA, volume 812, 2014. Communications Societies. Proceedings. IEEE, [11] B. Ferris, D. Fox, and N. D. Lawrence. Wifi-slam volume 2, pages 775–784. Ieee, 2000. using gaussian process latent variable models. In [3] A. Bose and C. H. Foh. A practical path loss model for IJCAI, volume 7, pages 2480–2485, 2007. indoor wifi positioning enhancement. In Information, [12] S. Gezici. A survey on wireless position estimation. Communications & Signal Processing, 2007 6th Wireless personal communications, 44(3):263–282, International Conference on, pages 1–5. IEEE, 2007. 2008. [4] M. Brunato and R. Battiti. Statistical learning theory [13] M. C. Gonzalez, C. A. Hidalgo, and A.-L. Barabasi. for location fingerprinting in wireless lans. Computer Understanding individual human mobility patterns. Networks, 47(6):825–845, 2005. Nature, 453(7196):779–782, 2008. [5] N. Bulusu, J. Heidemann, and D. Estrin. Gps-less [14] I. . W. Group et al. Wireless lan medium access low-cost outdoor localization for very small devices. control (mac) and physical layer (phy) specifications. IEEE personal communications, 7(5):28–34, 2000. 1997. [6] A. Cayley. On a theorem in the geometry of position. [15] Y. Gu, A. Lo, and I. Niemegeers. A survey of indoor Cambridge Mathematical Journal, 2:267–271, 1841. positioning systems for wireless personal networks. [7] N. Chang, R. Rashidzadeh, and M. Ahmadi. Robust IEEE Communications surveys & tutorials, indoor positioning using differential wi-fi access points. 11(1):13–32, 2009. IEEE Transactions on Consumer Electronics, [16] B. Hofmann-Wellenhof, H. Lichtenegger, and 56(3):1860–1867, 2010. J. Collins. Global positioning system: theory and [8] M. J. De Smith, M. F. Goodchild, and P. Longley. practice. Springer Science & Business Media, 2012. Geospatial analysis: a comprehensive guide to [17] K. Kaemarungsi and P. Krishnamurthy. Modeling of principles, techniques and software tools. Troubador indoor positioning systems based on location Publishing Ltd, 2007. fingerprinting. In INFOCOM 2004. Twenty-third [9] F. Evennou and F. Marx. Advanced integration of wifi AnnualJoint Conference of the IEEE Computer and and inertial navigation systems for indoor mobile Communications Societies, volume 2, pages positioning. Eurasip journal on applied signal 1012–1022. IEEE, 2004. processing, 2006:164–164, 2006. [18] C. Kang, S. Gao, X. Lin, Y. Xiao, Y. Yuan, Y. Liu, [10] R. Faragher and R. Harle. An analysis of the accuracy and X. Ma. Analyzing and geo-visualizing individual of bluetooth low energy for indoor positioning human mobility patterns using mobile call records. In 2010 18th International Conference on human mobility in big event settings using bluetooth. Geoinformatics, pages 1–7. IEEE, 2010. In Proceedings of the 3rd ACM SIGSPATIAL [19] N. E. Klepeis, W. C. Nelson, W. R. Ott, J. P. international workshop on indoor spatial awareness, Robinson, A. M. Tsang, P. Switzer, J. V. Behar, S. C. pages 51–58. ACM, 2011. Hern, W. H. Engelmann, et al. The national human [34] F. Thomas and L. Ros. Revisiting trilateration for activity pattern survey (nhaps): a resource for robot localization. IEEE Transactions on Robotics, assessing exposure to environmental pollutants. 21(1):93–101, 2005. Journal of exposure analysis and environmental [35] W. R. Tobler. A computer movie simulating urban epidemiology, 11(3):231–252, 2001. growth in the detroit region. Economic geography, [20] Y.-S. Kuo, P. Pannuto, K.-J. Hsiao, and P. Dutta. 46(sup1):234–240, 1970. Luxapose: Indoor positioning with mobile phones and [36] J. Torres-Sospedra, J. Avariento, D. Rambla, visible light. In Proceedings of the 20th annual R. Montoliu, S. Casteleyn, M. Benedito-Bordonau, international conference on Mobile computing and M. Gould, and J. Huerta. Enhancing integrated networking, pages 447–458. ACM, 2014. indoor/outdoor mobility in a smart campus. [21] B. Li, T. Gallagher, A. G. Dempster, and C. Rizos. International Journal of Geographical Information How feasible is the use of magnetic field alone for Science, 29(11):1955–1968, 2015. indoor positioning? In Indoor Positioning and Indoor [37] Y. Wang, X. Jia, H. Lee, and G. Li. An indoors Navigation (IPIN), 2012 International Conference on, wireless positioning system based on wireless local pages 1–9. IEEE, 2012. area network infrastructure. In 6th Int. Symp. on [22] T.-N. Lin and P.-C. Lin. Performance comparison of Satellite Navigation Technology Including Mobile indoor positioning techniques based on location Positioning & Location Services, number 54, 2003. fingerprinting in wireless networks. In 2005 [38] N. E. Williams, T. A. Thomas, M. Dunbar, N. Eagle, international conference on wireless networks, and A. Dobra. Measures of human mobility using communications and mobile computing, volume 2, mobile phone records enhanced with gis data. PloS pages 1569–1574. IEEE, 2005. one, 10(7):e0133630, 2015. [23] H. Liu, H. Darabi, P. Banerjee, and J. Liu. Survey of [39] C. Wu, Z. Yang, Y. Xu, Y. Zhao, and Y. Liu. Human wireless indoor positioning techniques and systems. mobility enhances global positioning accuracy for IEEE Transactions on Systems, Man, and mobile phone localization. IEEE Transactions on Cybernetics, Part C (Applications and Reviews), Parallel and Distributed Systems, 26(1):131–141, 2015. 37(6):1067–1080, 2007. [40] Y. Xu, S.-L. Shaw, Z. Zhao, L. Yin, Z. Fang, and [24] J. Liu, R. Chen, L. Pei, R. Guinness, and Q. Li. Understanding aggregate human mobility H. Kuusniemi. A hybrid smartphone indoor patterns using passive mobile phone location data: a positioning solution for mobile lbs. Sensors, home-based approach. Transportation, 42(4):625–646, 12(12):17208–17233, 2012. 2015. [25] D. E. Manolakis. Efficient solution and performance [41] Z. Yang and Y. Liu. Quality of trilateration: analysis of 3-d position estimation by trilateration. Confidence-based iterative localization. IEEE IEEE Transactions on Aerospace and Electronic Transactions on Parallel and Distributed Systems, Systems, 32(4):1239–1248, 1996. 21(5):631–640, 2010. [26] R. Mautz. Overview of current indoor positioning [42] M. A. Youssef, A. Agrawala, and A. U. Shankar. Wlan systems. Geodezija ir kartografija, 35(1):18–22, 2009. location determination via clustering and probability [27] E. Mok and G. Retscher. Location determination using distributions. In Pervasive Computing and wifi fingerprinting versus wifi trilateration. Journal of Communications, 2003.(PerCom 2003). Proceedings of Location Based Services, 1(2):145–159, 2007. the First IEEE International Conference on, pages [28] A. Motley and J. Keenan. Personal communication 143–150. IEEE, 2003. radio coverage in buildings at 900 mhz and 1700 mhz. [43] Y. Yue, T. Lan, A. G. Yeh, and Q. Li. Zooming into Electronics Letters, 24(12):763–764, 1988. individuals to understand the collective: A review of [29] E. Nordstr¨om, C. Diot, R. Gass, and P. Gunningberg. trajectory-based travel behaviour studies. Travel Experiences from measuring human mobility using Behaviour and Society, 1(2):69–78, 2014. bluetooth inquiring devices. In Proceedings of the 1st [44] P. A. Zandbergen. Accuracy of iphone locations: A international workshop on System evaluation for comparison of assisted gps, wifi and cellular mobile platforms, pages 15–20. ACM, 2007. positioning. Transactions in GIS, 13(s1):5–25, 2009. [30] K. Pahlavan, X. Li, and J.-P. Makela. Indoor [45] Y. Zheng, Q. Li, Y. Chen, X. Xie, and W.-Y. Ma. geolocation science and technology. IEEE Understanding mobility based on gps data. In Communications Magazine, 40(2):112–118, 2002. Proceedings of the 10th international conference on [31] T. Roos, P. Myllym¨aki, H. Tirri, P. Misikangas, and , pages 312–321. ACM, 2008. J. Siev¨anen. A probabilistic approach to wlan user location estimation. International Journal of Wireless Information Networks, 9(3):155–164, 2002. [32] J. S. Seybold. Introduction to RF propagation. John Wiley & Sons, 2005. [33] H. Stange, T. Liebig, D. Hecker, G. Andrienko, and N. Andrienko. Analytical workflow of monitoring