Supporting First Responders Localization During Crisis Management Massimo Ficco∗

Francesco Palmieri and Aniello Castiglione

Department of Industrial and Information Engineering, Second University of Naples [email protected]

Department of Computer Science, University of Salerno [email protected], [email protected]

Abstract—In recent years, homeland security has represented one of the most relevant application contexts, which has collected the interest of both public authorities and research community. In case of a crisis event, it is the responsibility of public and government authorities to manage the response operations. One of the constant challenges encountered when managing emergency is the lack of real-time location-aware information by the first responders that act on the crisis site. This paper aims at drawing an overall picture of the achieved research progressed for supporting localization of first responders during crisis events occurring in urban areas and involving critical infrastructures. In particular, this paper presents a hybrid positioning approach, which combines the strength of signals received from landmark nodes placed by first responders on the crisis site with information achieced from motion sensors. Finally, we investigate the key research topics in the considered area and the potential impact on the future developments scenario. Index Terms—Homeland security; crisis management; location-awareness; indoor positioning; pedestrian. The nal publication is available at: http://ieeexplore.ieee.org/document/7284940/

I. I NTRODUCTION Supporting of First Responders (FRs) represents one of the most critical activities during crisis events (terrorist attacks, disasters, and catastrophes). The modern society is faced with an increasing number of crises events, such as September 11, 2001, hurricanes in the Gulf of Mexico, earthquakes in Haiti, and the recent nuclear incident in Japan generated by the combination of a tsunami and an earthquake. Such events point out the vulnerabilities that exist in urban areas and in critical infrastructures. One of the main requirement in case of such events is the real-time availability of relevant location-aware information for the command & control centers, in order to perform the necessary analysis and to coordinate relief efforts. The damage (or worst, the loss) of communication capabilities together with the poor efficiency of positioning systems, especially in indoor environments, are the main current drawbacks. Therefore, during rescue operations, there is a gap among the situation of forces on the ground (e.g. police, firefighters etc.), the partial overview at Mobile Emergency Operation Centers (MEOC), and the overall overview at Command Emergency Operation Center (CEOC), i.e., the headquarter. Positioning of devices and people within indoor environments has been ∗ Corresponding author: Massimo Ficco, Second University of Naples, via Roma 29, I-81031 Aversa, Italy. [email protected].

the subject of many research and development efforts [1]. The interworking between the Wireless Wide Area communication networks (e.g., UMTS, HSDPA), and the current generation of Wireless Local Area Networks (e.g., Wi-Fi), has allowed to leverage wireless networks for provisioning of locationbased services [2]. The spreading of wireless hot-spots into public and private places, and the availability of positioning solutions based on Wireless Sensor Networks (WSN), as well as the current generation of mobile devices supporting several technologies used for localization [3], have fostered the development of new indoor positioning systems based on preinstalled (fixed) infrastructures [4], [5]. On the other hand, during a crisis or disaster, the pre-installed landmarks or anchors nodes could not be operational in the involved sites, and the training data needed to calibrate the positioning systems could not be available, fostering the need for landmark-free systems, that can perform self-localization without relying on any external landmarks. However, in landmark-free systems errors may be accumulated due to sensor noise, if no landmarks are available for recalibration. So, at the state of art, there are no commercially available positioning system that can be reliably used by FRs in the contexts in which they operate. For this reason, it is mandatory to investigate solutions for combining the landmark-based and landmark-free technologies, according to the danger area in which FRs have to operate. In this paper, we explore an effective solution for acquiring the location-aware information necessary to support the FRs in crisis areas, in which no assumptions can be done about the working conditions because of unpredictable events that can affect the availability of fixed positioning infrastructures. Thus, at first we present a short survey limited in scope to a subset of research works whose goal is supporting location-awareness for FRs. Then, we present a hybrid positioning approach that combines the strength of signals received from landmark nodes, which have to be manually placed by FRs within the crisis area (e.g., at the building entrance and along a flight of stairs), and the information gathered from motion sensors, such as gyroscope, accelerometer and compass available on the field. Specifically, landmarks are used as navigational support through relative positioning, whereas motion sensors can be used for inferring the action of the FR (such as speed and orientation), in order to achieve a more accurate positioning in a quite short time. Finally, some discussions and research

perspectives in the involved scenario are presented. II. C ONTEXT AND RELATED WORK A. Emergency management during crisis events A generic operational scenario during crisis management is represented in Fig. 1. FRs that act on the sites of interest must be enabled to collaborate and collect location-aware sensible data (e.g., temperature, chemical information), which has to be sent to the different MEOCs that manage the local activities. Only the CEOC has a global vision, so that it can coordinate the MEOCs activities by issuing the proper commands. The management of such scenario requires a suite of technologies able to provide real-time location-aware information and communication support to FRs. Typically, different networks are involved: • Personal Area Networks (PANs): ad-hoc networks in which a number of sensory data sources (used to acquire context-aware information) are arbitrarily connected to a smaller number of mobile devices (associate to the FRs); • Incident or Event Area Network (IAN): a network deployed for the disaster occasion, allowing person-toperson communication and forming a logical cell; • Jurisdiction Area Network (JAN): a communication network constituted by the proper devices installed in a specific jurisdiction to form a fixed or mobile infrastructure providing services associated to the jurisdiction; and • Extended Area Network (EAN): a network being able to support wide areas like national territories. B. Related work Localization systems can be classified into two main categories: Landmark-based and Landmark-free, depending on what kind of sensor devices are used. Landmark-based systems rely on certain proximity measurements between a mobile device and multiple landmarks that are deployed in the involved environment [6], [7]. Specifically, fingerprint-based approaches require a preliminary site survey over the areas of interests in order to build the fingerprint database. Several efforts have been made to reduce this mapping effort, for instance, by performing measurement at a coarser, room-level granularity [8]–[10]. However, the overall pre-deployment effort remains substantial. Also if the ideas from Dair et al. Bahl05 make the need for mapping unnecessary, they assume the presence of a very dense deployment of WiFi transmitters, that is not common in typical WiFi installations. Therefore, the considerable manual cost and efforts, in addition to the inflexibility to environment dynamics are the main drawbacks of fingerprint-based methods. ModelBased approaches are based on signal propagation properties and access point locations, and use geometrical models to figure out locations [12]. The advantage of using these approaches is that they reduce the the measurement efforts compared to fingerprinting schemes, at the cost of decreased localization accuracy, due to the irregular signal propagation in indoor environments. Since RF propagation characteristics

may widely vary, the model parameters would have to be estimated specifically for each indoor space involved. Moreover, they still depend on knowledge of the landmark locations. However, accurately tracking mobile devices by using received signal strength (RSS) is a challenging task since RSS values are affected by a non-negligible noise in a complex indoor environment due to attenuation, shadowing and multipath effects. Moreover, during a crisis or disaster, the landmark positions can be unknown, and the training data needed for calibration can be not available. Alternatively, infrastructure-less solutions can be adopted. In particular, as described earlier, the landmark nodes can be manually deployed by the FRs, and used as navigational support through relative positioning and path-based navigation. For example, FRs can move along a line of nodes in order to find the way to the exit or a specific location. Landmark-free approaches use the action sequences inferred from compass and accelerometer, and reconstruct the location trajectory, for example, via semi-supervised manifold learning techniques [13]. However, in Landmark-free systems errors may be accumulated due to sensor noise if no landmarks are available for re-calibration. Therefore, it is necessary to address the problem by using an hybrid positioning approach that combines Landmark-based and Landmark-free systems, as previously asserted. III. T HE H YBRID P OSITIONING A PPROACH The presented positioning approach is essentially based on a multi-sensor assisted pedestrian navigation scheme. It combines RSS- fingerprinting with pedestrian dead reckoning (PDR) algorithms. Moreover, 3D positioning can be easily achieved by adding relative height information (by barometer measurements). An appropriate Kalman filtering technique is used for sensor information fusion. Specifically, the proposed approach assumes that a fixed wireless infrastructure does not exist in the crisis area. During a mission, FRs deploy a limited set of landmarks nearby specific places (e.g., at the building entrance and along a flight of stairs, close to the lift), which can be used as reference points. Once deployed, known their placement and a partial knowledge of the floor-plan, the system is able to estimate an approximate radio map (RM) of the indoor space, which can be used to identify approximately the place in which the FR is moving. Therefore, we assume that FRs hold mobile devices and navigate in the crisis environment by relying on landmarks that work as reference points. Mobile devices can periodically send out beacon signals and measure RSS values from all the deployed landmarks, which can be used to estimate a coarse position (e.g., the room in which the mobile device is moving). Moreover, each mobile device has additional sensors (compass and accelerometer) for measuring the moving direction and speed of the FR, which can be used to estimate more accurate navigation information. A. Landmark-based Position Inference Assuming that during an emergency, FRs do not have the RM of the environment in which they will operate, we

Fig. 1.

Emergency scenario

propose a solution for simplifying the calibration process of the positioning infrastructure (landmarks) deployed by FRs in the crisis area. Specifically, known the description of the involved area and the position of placed landmarks, the proposed solution allows MEOC to compute at run-time a RM of the indoor space, which can be used by the FRs’ mobile devices to estimate a coarse indoor position. In our previous work [17], we presented a framework which allows to describe the characteristics of the considered area (in terms of hallways, rooms, walls, obstacles), by means of a graphical tool. Moreover, by indicating the location in which the landmarks are placed, the framework computes the RSS RM of the considered environment, by using a Multi-wallfloor (MWF) propagation model. It models the path loss (PL) between a receiver and a transmitter, located at a distance d, according with Equ. (1). P L (d)dB = 10 log 10 log

(4π)2 d20 Gt Gr λ2

Pt Pr

⇒ (1)

+ 10 log

d2 d20

+ 10 log β

where d0 is the distance from the transmitter, (Pt , Gt ) and (Pt , Gt ) are respectively the transmitter and receiver power and antenna gain, whereas λ is the wavelength, and β gathers all the attenuation factors met along the propagation path towards the receiver. Equ. (2) specializes the factor βdB according with Equ. (3), in which kwi and Lwi are respectively the number and the loss coefficients associated to the obstacles of type i, such as

walls, doors, windows, and so on, whereas kfl and Lfl denote respectively the number and the loss coefficients of the floors. The values for Lwi and Lfl can be obtained by the literature ( [14]).   d2 d20

P L (d)dB = (d0 )dB +

βdB =

M X

kwi Lwi +

i=1

F X

+ βdB ,

(2)

dB

kfl Lfl .

(3)

l=1

Considering the large noise that affects the wireless signal in a indoor environment (due to refraction, reflection, and diffraction phenomena, caused by the presence of absorption structures and human bodies), the use of a MWF for 3D positioning in lack of a rigorous training phase (the RSS site survey process) can only provide a low accuracy. Thus, we adopt a 2D propagation model and the barometric measuremens for the height estimation. Therefore, if the analysis is conducted on a single floor, the loss coefficients related to the floors can be neglected, and Equ. (2) can be rewritten by Equ. (4), to which is referred as Multi-Wall Classic (MWC) approach.  P L (d)dB = (d0 )dB +

d2 d20

 + dB

M X

kwi Lwi .

(4)

i=1

Moreover, assuming that the received signal strength prediction is strictly connected to the path loss estimation, in order to estimate the RSS average, Equ. (4) can be manipulated and reformulated in order to obtain the RSS expressed by Equ.

(5):
µPr µd , µLwi = Pt



Gt Gr λ 4πd0

2



d0 µd

2

1 . (5) i=1 kwi µLwi

QM

Equ. (5) provides the average value of Pr under the assumption that the transmitted power, the number of obstacles kwi , the receiver and transmitter antenna gain are known and invariant along the time. The distance d and the loss coefficients are modeled as stochastic processes, of which µd , σd2 , and µLwi , 2 σL are their first-order moments (i.e., mean and variance). wi However, the MW approach exploits a discretization of the environment in a grid of squared cells, each one of dimension equal to [qX; q(X + 1)] x [qY ; q(Y + 1)], where q denotes for the cell quantization step, whereas X and Y are the indexes identifying the coordinates of the cell in the grid. The grid has to fully cover the indoor environment involved into the analysis. Each cell is characterized by an inside uniform distribution of signal strength referred to its center (i.e., at an average distance d from a transmitter) provided by Equ. (6): µR =

q (x − q(2XX + 1)/2)2 + (y − q(2Y Y + 1)/2)2 .

(6)

2 This discretization process introduces an uncertainty σR in the distance estimation, whose value can be straightforwardly estimated with q 2 /3. However, it is not the only factor that contributes to the uncertainty of the MW approach. In particular, stemming from the guideline figured out by [15], the uncertainty can be achieved by applying the law of propagation of uncertainty in the case of uncorrelated variables, which can be expressed by Equ. (7):

by means of the MWF based model, a deterministic position inference method is adopted. In particular, the most common algorithms adopted to infer location use the Euclideanbased method to compute distance between the measured RSS sample and each RSS fingerprint in the RM ( [16]). The coordinates associated with the vector in the RM that provides the smallest Euclidean distance is returned as the position estimation. Specifically, during the operational phase, in order to estimate the user position, the RSS sample collected from the landmarks is compared with all the existing RSS fingerprints in the computed RM. The metric used to make such comparison is the signal distance between the two RSS vectors, expressed as follows ( [16]): v v unAP unAP uX uX 2 t (ssk − fi,k ) = t d2i,k , ∆= k=1

where the vector SS = [ss1 ,ss2 ,...,ssnLK ] is the RSS sample measured at current FR location from nLK landmarks, whereas the vector Fi = [fi,1 ,fi,2 ,...,fi,nLK ] represents the set of RSS values computed by the proposed approach. The fingerprint entry Fi that has the closest match to the collected SS sample (i.e., the one exposing the lowest distance di,k ) is used by the system as the estimation of the current FR location. The si element is assumed to be the mean of the n RSS measurements. Finally, we have evaluated the quality of the presented MWC-based fingerprinting approach in an actual case scenario, by using the following two metrics: •

σPr

v   2 u u ∂Pr 2 ∂Pr =t σd2 + ∂d d=µD ∂Lwj L

2 . σL wj

(7)

wj =µLw j

Equ. (7) shows that the standard deviation of MWC model depends on these two main contributors. The former is connected to the quantization step, which can be used to modify the value in the model, whereas the latter relies on the capability of measuring the loss coefficient of the obstacles with high precision. Summarizing, the average and standard deviation achieved through the MWC approach are respectively expressed by Equ. (5) and Equ. (7). They are used in the proposed approach to compute the quality of the computed RM. A variation of the MWC approach is the linear one (MWL), in which the attenuation factor is proportional to an α coefficient (i.e., linearly dependent from d) as represented by Equ. (8):  P L (d)dB = (d0 )dB +

αd2 d20

 + dB

M X

kwi Lwi .

(8)

i=1

In order to simplify, in this work we used a MWL model with α equal to 3.6, obtained after a series of measurements within the experimental environment and in accordance to the suggestions provided by [18]. Finally, in order to evaluate the quality of the RM computed

(9)

k=1

Accuracy (α) is the degree of closeness of the estimated position δ(cal,i) to the actual (true) user location δact . It is the mean error computed as: α=

•

K 1 X (δ(calc,i) − δactual ). K i=1

(10)

Precision (ρ) is the percentage of runs where the estimated position differs from the true location by less than a fixed accuracy αT : PK 1 ρ= K j=1 (Θ(δ(estim,j) , δactual )), with

Θ(δ(est,j) , δactual ) =

n

0 1

if δ(est,j) − δact ≥ αT if δ(est,j) − δact < αT

(11) For instance, a positioning system that provides an accuracy/precision level equal to 1.5m/75% achieves an accuracy of 1.5m, with at most a probability of 75%. Moreover, we modeled the environment as a space composed of square areas, each representing a grid’s cell of RxR meters. Several measurement campaigns have been performed by varying the size of the cells. Tab. I shows the likelihood of returning the correct location (precision) with respect to the grid spacing, which can be assumed equal to the accuracy of the positioning system.

TABLE I P RECISION VERSUS THE GRID SPACING

Precision (%) 73 81 86 89

Accuracy (m) 1 2 3 4

B. Pedestrian Navigation As described in the previous section, the MWC-based fingerprinting allows to infer coarse positions with an accuracy of the order of two 2 meter and a precision of about 80%. The provided quality of positioning can be considered acceptable for supporting FRs in a crisis scenario. In general, it may be sufficient to know only the room in which the FR is located. On the other hand, for a finer localization, it can be necessary to know how the FR is moving in the room, or however, within the grid’s cell. Thus, inertial systems theory yields a corrector for the state prediction. It can be used to estimate relative motion (with respect to the center of the grid’s cell) over short intervals. As relative positioning solution, a foot-mount inertial pedestrian navigation (IMU) is adopted. Moreover, a step-based pedestrian dead reckoning (PDR) algorithm is used to estimate step detection, step length and step direction [19]. The step detection is based on pedestrian gait cycle, which includes the following sequential phases: push-off, swing, heel strike and stance [20]. The gait cycle phases can be detected from acceleration. The stride length of pedestrian walking can be estimated by using personal constants and a linear relationship with the detected frequency [21]. The step direction determination is estimated by gyroscope measurements. Finally, in order to identify changes in the height a pressure measurement p from barometer can be converted to altitude information h by the following equation: L ∗ h gM p = p0 (1 − ) RL . T0

(12)

The parameters can be assumed constant for standard atmospheric conditions and their values are provided in [22]. The method for PDR is based on an Adaptive Kalman Filtering (AKF)-based framework described in [19]. It is used to estimate the errors which accumulate due to the IMU drifts. The AKF is updated with velocity measurements and angular rates by the Zero-Velocity-Update strategy (ZUPT) every time the foot is on the floor [23]. The ZUPT resets the velocity to zero each time the foot comes to rest on the ground, and thus, limits the error growth of the system to a linear function. Fig. 2 shows the main blocks of the methodology. C. Hybrid Navigation Model The implemented navigation model predicts the new position based on the knowledge of the previous one, estimated by an MWC-based approach. Specifically, it is a hybrid model that uses RSS-based fingerprinting, the barometric height, and

Fig. 2.

Kalman-based approach used for pedestrian dead reckoning.

IMU measurements to infer the measurement vector needed for positioning estimation. The navigation model can be expresses as follows: 

  px (i) ωx (i)  py (i)  =  ωy (i) pz (i) hz (i − 1)

+ +

 rx (i) ry (i)  ,

(13)

where the vector [px , py , pz ] represents the FR position, (ωx , ωy ) is the 2D position estimated by the MWC-based fingerprinting, (rx , ry ) is the relative movement with respect (ωx , ωy ). The relative movement (rx , ry ) is estimated by the IMU according to Equ. 14: 

rx (i) ry (i)



 =

r(i − 1) r(i − 1)

+ +

sx (i)cosθ sy (i)senθ

 + (i),

(14)

where s represents the estimated stride length, θ is the estimated heading of the user, whereas  is the system process noise vector. IV. E XPERIMENTAL A NALYSIS For the experimental campaign, we adopted the widely used Xsens MTw foot mounted IMU, which was connected to the PC via wireless (IEEE 802.15.4). IMU sends data with a rate of 120 Hz. The wireless sensors adopted for the experimental evaluation consist of a PINGUINO-OTG (PIC32MX440-F256H) 80 Mhz microcontroller with a ZigBee MRF24J-40MA 2.4 GHz IEEE 802.15.4 radio transceiver module. Specifically, they act either as landmarks, if they were configured in slave-mode, or as readers connected to the PC, if they were configured in master-mode. Fig. 3 shows the hardware system. In order to evaluate the proposed approach, in Fig. 4 is shown an example of trajectory on two floors, which is

In Fig. 4 are represented by red points, the positions estimated by using only MWC-based fingerprinting. The provided results show an accuracy of about 1m with a precision of 69%. On the other hand, by using the IMU, it is possible to estimate the relative movement with the respect to the last MWC-based estimation, as well as the trajectory following by the FR in the grid’s cell.

Fig. 3.

Experimental system.

recorded inside our institute building, which is considered as a typical office building. The trajectory starts on the first floor. The person moves the stairs up to the second floor, walks around and goes back downstairs to the starting point. The map of each floor has been organized as a grid cells of dimension of 1x1 meters. We simulated the movement of a person along the considered trajectory and the IMU is mounted on the foot. Moreover, in the considerate scenario, we assume that are placed three landmarks per floor.

Fig. 5.

Positioning by the hybrid approach.

Finally, in Fig. 6 is shown the position estimation in zdirection drifts over time. It shows the estimated height over time starting with an initial value of about 0.5m on the first floor. Figure shows that the person walked up to the second floor, resulting in a change of about 4m, which corresponds to the height of the second floor.

Fig. 6. Fig. 4.

Positioning by MWC-based approach.

Height information from barometer.

V. C ONCLUSIONS AND FUTURE WORK In case of crisis events, it is the responsibility of public authorities to manage the response operations in order to save lives and restore a sense of order. One of the constant challenges encountered by public authorities when managing emergency situations is the lack of actionable location-aware information, i.e., the information which is required for making fast and correct decisions under pressure. Therefore, in this paper we presented a hybrid positioning approach, which combines the landmark-based and landmark-free technologies. The indoor localization in danger area involves a very specific setting, which is unique and has not been studied extensively enough in the literature. Specifically, the design of navigation systems implies the wicked problem of having to create ad-hoc technologies to support navigation, and at the same time, to create navigation practices upon the created technology. With the technology readily available today, it is possible to capture movement and additional forms of environmental data, which can prove to be useful for supporting navigation. Therefore, the main issue to be addressed in future work will be: how does the designed system enrich the context of FRs, and how this enrichment supports the construction of navigation practices? R EFERENCES [1] Chenshu Wu, Zheng Yang, Yunhao Liu, and Wei Xi, WILL: Wireless Indoor Localization without Site Survey, in IEEE Trans. on Parallel & Distributed Systems, vol. 24, no. 4, Apr. 2013, pp. 839-848. [2] M. Ficco, R. Pietrantuono, and S. Russo. Supporting ubiquitous location information in interworking 3G and wireless networks, in Communications of the ACM, vol. 53, no. 11, 2010, pp. 116-123. [3] M. Ficco and S. Russo. A Hybrid Positioning System for Technologyindependent Location-aware Computing, in Software: Practice and Experience, vol. 39, Feb. 2009, pp. 1095-1125. [4] M. Ficco, P. Palmieri, and A. Castiglione. Hybrid indoor and outdoor location services for new generation mobile terminals, in Personal and Ubiquitous Computing, vol. 18, no. 2, Feb. 2014, pp. 271-285. [5] R. Aversa, B. Di Martino, M. Ficco, and S. Venticinque. A simulation model for localization of pervasive objects using heterogeneous wireless networks. in Journal of Simulation Modelling Practice and Theory, vol. 19, no. 8, Sep. 2011, pp. 1758-1772. [6] B. Ferris, D. Fox, and N. Lawrence. WiFi-SLAM using gaussian process latent variable models, in Proc. of the 20th Int. Joint Conf. on Artificial Intelligence, 2007, pp. 24802485. [7] T. Yairi. Map building without localization by dimensionality reduction techniques, in Proc. of the 24th Int. Conf. on Machine Learning, 2007, pp. 10711078.

[8] A. Haeberlen, E. Flannery, A. M. Ladd, A. Rudys, D. S. Wallach, and L. E. Kavraki. Practical Robust Localization over Large-Scale 802.11 Wireless Networks, in Proc. of the 10th Int. Conf. on Mobile Computing and Networking, 2004, pp. 70-84. [9] M. Ficco, C. Esposito, and A. Napolitano. Calibrating Indoor Positioning Systems With Low Efforts, in IEEE Transactions on Mobile Computing, vol. 13, no. 4, Apr. 2014, pp. 737-751. [10] C. Esposito and M. Ficco. Deployment of RSS-Based Indoor Positioning Systems. Journal of Wireless Information Networks, vol. 18, no. 4, Dec. 2011, pp. 224-242. [11] P. Bahl, J. Padhye, L. Ravindranath, M. Singh, A. Wolman, and B. Zill. DAIR: A Framework for Managing Enterprise Wireless Networks Using Desktop Infrastructure. in Proc. of the Hot Topics in Networks Workshops, 2005. [12] D. Maligan, E. Elnahrawy, R. Martin, W. Ju, P. Krishnan, and A. S. Krishnakumar. Bayesian indoor positioning systems, in Proc. of the 24th IEEE Int. Conf. on Computer Communications Societies, 2005, pp. 12171227. [13] J.J. Pan, S.J. Pan, J. Yin, L.M. Ni, and Q.Y. Fellow. Tracking Mobile Users in Wireless Networks via Semi-Supervised Co-Localization, in IEEE Tran. on Pattern Analysis and Machine Intelligence, vol. 34, no. 3, Mar. 2012, 587-600. [14] H. Hashemi. The Indoor Radio Propagation Channel, in Proc. of the IEEE, vol. 81, no. 7, Jul. 1993, pp. 943-968. [15] Int. Organization for Standardization, Guide to the expression of uncertainty in measurement (GUM)-Supplement 1: Numerical methods for the propagation of distributions, ISO draft guide DGUIDE99998, 2013, available at: http://www.bibsonomy.org/bibtex/2eea813f6ec9a5d910a87439a5b453a19/eutrimer [16] M. Youssef and A. Agrawala. Handling Samples Correlation in the Horus System, in Proc. of the IEEE INFOCOM, 2004, pp. 1023-1031. [17] M. Ficco. Calibration-less indoor location systems based on wireless sensors, in Journal of Ambient Intelligence and Humanized Computing, vol. 5, no. 2, April 2014, pp. 249-261. [18] K. Kaemarungsi and P. Krishnamurthy. Modeling of indoor positioning systems based on location fingerprinting, in Proc. of the IEEE INFOCOM, vol. 2, 2004, pp. 1012-1022. [19] W. Chai, C. Chen, E. Edwan, and J. Zhang. 2D/3D Indoor Navigation Based on Multi-sensor Assisted Pedestrian Navigation in Wi-Fi Environments, in Proc. of the Int. Conf. on Ubiquitous Positioning, Indoor Navigation, and Location Based Service, 2012, pp. 1-7. [20] S. Godha and G. Lachapelle. Foot mounted inertial system for pedestrian navigation, in Measurement Science and Technology, vol. 19, no. 7, 2008, PP. 1-9. [21] X. Zhao, Z. Syed, B. Wright and N. El-Sheimy. An Economical and Effective Multi-sensor Integration for Portable Navigation System, in Proc. of the 22nd Int. Technical Meeting of the Satellite Division of the Institute of Navigation, 2009, pp. 2088-2095. [22] T. Gadeke, J. Schmid, M. Zahnlecker, W. Stork. Smartphone Pedestrian Navigation by Foot-IMU Sensor Fusion, in Proc. of the Int. Conf. on Ubiquitous Positioning, Indoor Navigation, and Location Based Service, 2012, pp. 1-8. [23] E. Foxlin. Pedestrian tracking with shoe-mounted inertial sensors, in IEEE Computer Graphics and Applications, vol. 25, no. 6, 2005, pp. 38-46.