Orientation-based Wi-Fi Positioning on the Google Nexus One

Transcription

1 200 IEEE 6th International Conference on Wireless and Mobile Computing, Networking and Communications Orientation-based Wi-Fi Positioning on the Google Nexus One Eddie C.L. Chan, George Baciu, S.C. Mak Department of Computing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong Abstract While localization systems for indoor areas using the existing wireless local area network (WLAN) infrastructure have recently been proposed, wireless LAN localization approaches suffer from a number of significant drawbacks. To begin with, there is inaccurate position tracking due to the orientation of the mobile device and signal fluctuation. In this paper, we apply an orientation filter and a Newton (TR) algorithm to eliminate the noisy location estimation. We implement the localization algorithm on the Nexus One which is a Wi-Fi enabled device with a digital compass. The average error distance is only.82m. We achieve 90% precision within 2.45m. The proposed method leads to substantially more accurate and robust localization system. Index Terms Indoor positioning, Digital compass, Orientation, Newton, Location Fingerprinting at C i. The vector Y i is the fingerprint of the location C i. When a new fingerprint Y is derived from a transmitter at an unknown location A, we can locate A by searching for the fingerprint Y i that is closest to Y at d distance and estimate the location with the corresponding C i. The drawbacks of LF approach are needed to have extensive training dataset surveying and highly affected by the changing of internal building infrastructure, presence of humans and interference among devices leading to inaccurate localization. These issues have been addressed in our previous work [0], []. I. INTRODUCTION Over the last several years, positioning systems have been widely developed and many mobile devices have included positioning features. For example, Google s Nexus One employs a digital compass, Wi-Fi access and Global Positioning Systems (GPS). GPS is the most effective in relatively open and flat outdoor environments but is much less effective in nonline-of-sight (NLOS) environments such as hilly, mountainous, or built-up areas. Positioning systems for indoor areas [], [2] using the existing WLAN infrastructure have been suggested. Devices in WLAN are located using their signal strength, with the most typical approaches to locating WLAN-enabled devices being propagation-based and location-fingerprinting (LF) based. Propagation-based approaches measure the received signal strength (RSS), angle of arrival (AOA), or time difference of arrival (TDOA) of received signals and apply mathematical models to determine the location of the device. The drawbacks of propagation-based approaches [], [2], [3] are needed to compute every condition that can cause wave signal to blend in order to achieve accurate localization. LF-based approaches does not require to compute the wave phenomena to estimate the location. LF-based approaches [4], [3],[5], [6], [7], [8], [9] locate a device by comparing its coordinates with the received signal strengths (RSSs) and coordinates or other devices within the Wi-Fi footprint. These device coordinates are held in an LF database which is built up by collecting the data {(Y n,c i ),i =...N}, for N locations in an area, where C i is the known location of the i th measurement and Y i =(Y i,...,y in ) is the RSS vector when the transmitter is Fig.. User interface of the proposed positioning system on the Nexus One In our previous work we used the [2] and Newton (TR) algorithms [3] to enhance traditional LF by filtering the noisy signal. Although we have achieved to have 5% more accurate positioning than traditional LF approaches, this was not completely effective because they do not take into account the orientation of a user. Intuitively, it would be better to track the target object with the correct direction /0/$ IEEE 392

2 In this paper, we focus to enhance our previous work [3] and apply an orientation filter to previous localization algorithm. The proposed approach is in four phases. In the first phase we detect the IEEE 802.b Wi-Fi signal strength and collect the LFs into a training database. In the second phase, we retrieve the LFs (location fingerprints) and the direction using the Google Nexus One. Figure shows the user interface of our proposed positioning system on the Nexus One. In the third phase we make use of orientation filter to pick up the correct location fingerprint and estimate the location by probabilistic LF approach. Finally, TR method to adjust the location estimation. Our proposed method requires 0% fewer access points (APs) and is 7% more effective than our previous method. The rest of this paper is organized as follows: Section II presents the positioning methodologies. Section III presents experimental setup of WLAN tracking in a campus floor. Section IV discusses the performance evaluation of our experiment. Finally, Sections V offers our conclusion and future work. II. POSITIONING ALGORITHMS Similar with our previous works [0], [2], [3], we make use of LF approach to track a WLAN-enabled device, but this time we only pick up LF within the direction. Finally, we use TR algorithm (8)-(3) to adjust the location estimation. A. Probabilistic Location Fingerprinting Algorithm Probabilistic LF calculates the most probable location out of the pre-recorded LF database. We select a location d i if P (d i s) > P (d j s), for i, j =, 2, 3,...,n, j i. Using Bayes formula [4], and assuming that P (d i s) =P (d j s) for i, j =, 2, 3,...,n we have the following decision rule based on the likelihood that P (s d i ) is the probability that the signal vector s is received, given that the WLAN-enabled device is located in location d i. We can estimate d i by P (d i /S) = P (S/d i) P (d i ) P (S) Since P (S) is constant for all d, the algorithm can be rewritten: () P (d i /S) =P (S/d i ) P (d i ) (2) The estimated location d is the one which obtains the maximum value of the probability in equation (2): d = arg max [P (d i /S)] = arg max [P (S/d i ) P (d i )] (3) d i d i B. Orientation Filter Given the pitch (θ)and roll (φ) rotational angles, we can transform the magnetic components to the local level plane coordinate system, and then, determine the azimuth, or compass direction (γ), as follows [5]. X H = X cos (φ)+y sin (θ) sin (φ) Z cos (θ) sin (φ) Y H = Y cos (θ)+z sin (φ) γ = arctan (Y H /X H ) (4) where (X H, Y H ) are the horizontal components of the Earth s magnetic field. A valid location fingerprint sample is a sample that passes the filtering test based on the AP s one-hop and two-hop neighboring APs positions within the compass direction γ. However, when a user moves, there may be some error α between the measured orientation and the actual compass direction. In order to cover all possible directions, we need to define an orientation region S as follows: α S = u 2 dθ (5) 2 α where u is the maximum distance of two-hop neighboring APs positions. We apply the filter to this region to select the valid location fingerprint. The filter condition is as follows: { fi f i S, (6a) filter(f i )= 0 f i / S. (6b) C. Method For the completness, we include our previous work which use the Newton Trust-Region algoithm to derive the trajectory in a bounded region iteratively and they are more effective than other iterative non-linear optimization methods. TR method would be especially suitable for selecting iterates in the trajectory estimation process. Signal propagation loss algorithm [6][][2] calculates the received signal strength (RSS) with path loss as follows: R = r 0α log 0 (d) wallloss (7) where r is initial RSS, d is a distance from APs to a location, α is the path loss exponent (clutter density factor) and wallloss is the sum of the losses introduced by each wall on the line segment drawn at Euclidean distance d. Consider a typical unconstrained minimization of location error problem, min f(x) (8) x V where V is a vector space. f(x) is derived from the signal propagation loss algorithm in (7). At iteration k, with interate x k and TR radius k, the TR set is: A k = {x V x x k k k} (9) There are three calculations to be made in the TR method: ) Calculation of the TR subproblem where the goal is to approximate the location minimizer in the region, 2) Calculation of the TR fidelity, where the goal is to evaluate the accuracy of the location and 3) Calculation of the radius, to determine the size of a. 393

3 ) TR subproblem: A quadratic model m k is constructed to approximate f(x) within the TR. <, > denotes the inner product. The goal of a TR subproblem is to compute whether x k + s k is in the region. m k (x k + s) =m k (x k )+ g k,s + s, H x S (0) 2 where m k (x k )=f(x k ), g k is the gradient or first derivative of f(x) at x k, and H k is the Hessian of f or second derivative of f(x) at x k. When H K 0, m k is said to be a second-order model. A TR subproblem is then to compute an s k, the TR region subproblem is: s k = arg min ψ k (s) = g k,s + s, H x S () 2 2) TR fidelity: The trial point will be tested to see if it is a good candidate for the next iteration. This is evaluted by: p k = f (x k) f (x k + s k ) (2) m k (x k ) m k (x k + s k ) Suppose an initial is given and let η,η 2,γ, and γ 2 be some constants satisfying 0 <η <η 2 < and 0 <γ <γ 2 <. If p k η, then the trial point is accepted, i.e., x k+ = x k + s k. Otherwise, x k+ = x k. When m k approximates f well and yields a large p k, the TR radius will be expanded for the next iteration. Otherwise, if p k < η or p k < 0, m k does not approximates f well within the current region A k. Therefore, the iterate remains unchanged and the TR radius will be reduced to as to allow the derivation of a more appropriate model and subproblem for the next iteration. 3) TR radius: We can update the TR radius as follows: γ 2 k if p k η 2, k+ = k if p k [η,η 2 ], (3) γ k if p k <η where η and η 2 represent the lower and upper bound of TR fidelity. γ and γ 2 represent the changing ratio of the TR radius. The iterative process for (3) will be repeated until the sequence of iterates x k converges. III. EXPERIMENTAL SETUP In this section we investigate how TR method influences the accuracy with factors of number of APs, resolution and radius of. The experiment was established on the 7th floor of the PQ building, in Department of Computing, at The Hong Kong Polytechnic University. Figure 2 shows the floor plan of the 7th floor of the PQ building, at The Hong Kong Polytechnic University. In our experiment, we walked through the hallway on the 7th Floor with the Nexus One (our walking trajectory), using the proposed positioning algorithms to estimate the location of the device. The estimated and actual coordinates were calculated and collected at 30 locations in the hallway. The dimension of floor is approximately 50m by 20m. The received TABLE I SUMMARY OF EXPERIMENT SETUP IN A LABORATORY ENVIRONMENT Item Description Total laboratory area 000 square meter RSS variation Between -90 dbm and -35 dbm Number of sample points 4880 sample points Wi-Fi channel, 6, and Facing direction of each sampling North, South, East, and West signal sensitivity also limits the range of the RSS to be between -90 dbm and -30 dbm. Nevertheless, the highest typical value of the RSS is approximately -40 dbm at one meter from any AP. We use the same setting to measure RSS as used in [4], [3], [5], [6], [], [7], [8]. Table I summaries the experiment setup. IV. PERFORMANCE EVALUATION The major performance metrics of interest for WLAN localization are accuracy and precision. Accuracy considers the value of average distance errors. Precision considers how consistently the system works in its performance over many iterations. In the following, Subsection A presents the results of trajectory estimation. Subsection B describes our results for the number of access points relative to accuracy under the Orientation-based, [3], Kalman Filter [2] and traditional Location Fingerprinting method [0]. Subsection C describes the same for the relationship of resolution to accuracy (precision). Subsection D shows how our method reduces the error distance. A. Result for Trajectory Estimation Figure 2 shows the original and estimated walking trajectory on the 7th floor PQ building at The Hong Kong Polytechnic University under orientation-based,, and traditional Location Fingerprinting approach. As can be seen in Figure 2, due to signal fluctuation, the estimated path of traditional LF and bulged inside the room PQ703 and PQ77 sharply. approach depends on the current received signals to estimate the location. If signal fluctuates, then it will affect the accuracy of the localization. approach depends greatly on the priori and posteriori estimation to optimize the estimated location. Therefore, if the signal fluctuates, the posteriori estimation will be affected greatly, whereas orientation-based TR approach estimates more stably and locates more accurately. Figure 3 shows the squared error distance of estimated walking trajectory under Orientation-based, Trust Region, and Traditional Location Fingerprinting approach. Orientation-based TR approach eliminates the distance error better than TR, and LF approach. The mean squared errors of Orientation-based TR, TR, and LF approach are 3.33m, 4.0m, 4.78m and 6.8m respectively. 394

4 Start End Original Path Orientation-based Fig. 2. Original and estimated walking trajectory on the 7th floor PQ building at The Hong Kong Polytechnic University under orientation-based Trust Region,, and Traditional Location Fingerprinting approach Squared Error (m) Orientation based Fig. 3. The squared error distance under orientation-based, Trust Region, and Traditional Location Fingerprinting approach B. Effect of number of APs on the localization accuracy Figure 4 shows the relationship of number of access points to accuracy using each of the three methods. The resolution is 3m. We vary the number of APs from to 5. We choose APs by a first-come-first-get scheme. For example, if the number of APs sets to be 8, we only choose the first detected 8 APs. In general, there are three cases to operate the, enlarge, shrink and unchanged. First case, the will enlarge if the estimated point is outside and far away from the region. Second case, the orientation-based will shrink if the estimated point is near to the center of the region. Otherwise, the region will remain unchanged. A higher number of APs improves the accuracy but after more than 8 APs are used the accuracy does not increase significantly. The orientation-based TR method achieves from 50% to 80% accuracy when there is between 3 and 7 APs and becomes stronger as more APs are added. With 9 APs, it obtains 93% accuracy. We calculate and compare the total area of under each line in Figure 4. On average orientation-based TR method requires 5% and 20% fewer APs than and traditional LF method respectively to achieve effective localization. Perhaps the most important point to take away here is that neither the traditional LF approach nor the Kalman Filter at any point matches the accuracy of the orientationbased TR method and never achieve 90% accuracy. The accuracy of orientation-based orientation-based TR method have over 8% since the APs are 4, and after AP increase to 8, accuracy is over 92% and then increase slightly. C. Effect of resolutions on the localization accuracy Figure 5 shows the relationship of resolution to accuracy using the three methods. N meters resolution means we can locate a device correctly in a precision of N meters. Orientation-based TR radius is used similarly as last subsection. The orientation-based TR method achieves 74% accuracy with 2 meter resolution and 90% accuracy with 2.45 meters resolution. Orientation-based TR will almost certainly achieve 395

5 Accuracy Orientation based Number of APs of orientation-based TR method is.82m. And the orientationbased TR method reduces the average error distance from 2.49m (using the traditional LF approach) to 2.0m. The Orientation-based TR method can reduce the error distance from.73m to.38m on X-axis and.2m to.m on Y-axis. Interestingly, the error on the X-axis is greater than on the Y-axis (.38m and.m respectively). In the testing environment, the width of the floor (X-axis) is longer than the length of the floor (Y-axis) so it may be that along the longer X-axis there was more opportunity for the signal energy to be absorbed by air and furniture. We calculate and compare the total area of under each line in Figure 6. On average orientation-based TR method reduce 0% and 20% error to and traditional LF method respectively. In any case, these results show that the orientation-based TR method significantly reduces the error distance in position estimation. Fig. 4. Relationship of number of access points to accuracy under orientationbased,, and Traditional Location Fingerprinting approach Orientation based Error Distance (m) Accuracy Orientation based Resolution (m) Fig. 6. Error distance of position estimation under orientation-based Trust Region,, and Traditional Location Fingerprinting approach Fig. 5. Relationship of resolution(m) to accuracy under orientation-based,, and Traditional Location Fingerprinting approach perfect accuracy when the resolution is at 4.5m. D. Results for the error distance in trajectories In this section we consider the impact on the error distance when using the under orientation-based, Trust Region, and traditional Location Fingerprinting approaches. Figure 6 plots the error distance for position estimation. Figure 7 plots the error distance estimation on the X and Y axis. In this experiment, we select the location fingerprint by 80 degree which near the previous estimated location for estimating the position. The average error distance V. CONCLUSION In the future, wireless communications and mobility service provision will be characterized by global mobile access (terminal and personal mobility), a high quality of service with full coverage, and intelligible and simple access to multimedia services for voice and video via one user single terminal. Current approaches to WLAN localization suffer from signal fluctuations and the change of the direction, which lead to inaccurate tracking. In this paper, we apply an orientation filter and a Newton algorithm to eliminate the noisy location estimation. We test and compare it with our previous approaches. Our experimental analysis shows the effectiveness of proposed orientation-based Newton Trust Region method. This leads to substantially more accurate and robust localization system. Future work will be focused 396

6 Error Distance in X axis (m) Error Distance in Y axis (m) Orientation based Orientation based Fig. 7. Error distance of position estimation on the X and Y coordinates under Orientation-based,, and Traditional Location Fingerprinting approach on building a 3D pervasive tracking and a dynamic spatiotemporal filtering technique. [9] S. Fang, T. Lin, and P. Lin, Location Fingerprinting In A Decorrelated Space, Knowledge and Data Engineering, IEEE Transactions on, vol. 20, no. 5, pp , [0] C. L. Chan, G. Baciu, and S. C. Mak, Wireless Tracking Analysis in Location Fingerprint, 4th IEEE Wireless and Mobile Computing, Networking and Communications, WiMOB, pp , [] C. L. Chan, G. Baciu, and S. C. Mak, Fuzzy Topographic Modeling in WLAN Tracking Analysis, International Conference on Fuzzy Computation, [2] C. L. Chan, G. Baciu, and S. Mak, Using Wi-Fi Signal Strength to Localize in Wireless Sensor Networks, IEEE International Conference on Communications and Mobile Computing, CMC, pp , [3] C. L. Chan, G. Baciu, and S. C. Mak, Using the Newton Trust-Region Method to Localize in WLAN Environment, 5th IEEE Wireless and Mobile Computing, Networking and Communications, WiMOB, pp , [4] H. Liu, H. Darabi, P. Banerjee, and J. Liu, Survey of wireless indoor positioning techniques and systems, IEEE Transactions on Systems, Man, and Cybernetics, Part C: Applications and Reviews, vol. 37, no. 6, pp , [5] M. Caruso, Applications of magnetic sensors for low cost compass systems, Position Location and Navigation Symposium, IEEE 2000, pp , [6] K. Kaemarungsi and P. Krishnamurthy, Properties of indoor received signal strength for WLAN location fingerprinting, Mobile and Ubiquitous Systems - Networking and Services. MOBIQUITOUS. The First Annual International Conference, pp. 4 23, [7] W. Wong, J. Ng, and W. Yeung, Wireless LAN positioning with mobile devices in a library environment, 25th IEEE International Conference on Distributed Computing Systems Workshops, pp , [8] P. Bahl, V. Padmanabhan, and A. Balachandran, A Software System for Locating Mobile Users: Design, Evaluation, and Lessons, online document, Microsoft Research, February, ACKNOWLEDGMENT The authors would like to thank the Information Technology Service Center in The Hong Kong Polytechnic University to provide information about wireless infrastructure in The Hong Kong Polytechnic University. REFERENCES [] R. Jan and Y. Lee, An indoor geolocation system for wireless LANs, International Conference, Parallel Processing Workshops, pp , [2] P. Prasithsangaree, P. Krishnamurthy, and P. Chrysanthis, On indoor position location with wireless LANs, The 3th IEEE International Symposium, Personal, Indoor and Mobile Radio Communications., vol. 2, [3] J. Kwon, B. Dundar, and P. Varaiya, Hybrid algorithm for indoor positioning using wireless LAN, IEEE 60th Vehicular Technology Conference, VTC, vol. 7, [4] A. Taheri, A. Singh, and A. Emmanuel, Location fingerprinting on infrastructure 802. wireless local area networks (WLANs) using Locus, Proceedings of the 29th Annual IEEE International Conference on Local Computer Networks, pp , [5] K. Kaemarungsi and P. Krishnamurthy, Modeling of indoor positioning systems based on location fingerprinting, INFOCOM. Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies, vol. 2, [6] B. Li, Y. Wang, H. Lee, A. Dempster, and C. Rizos, Method for yielding a database of location fingerprints in WLAN, Communications, IEE Proceedings-, vol. 52, no. 5, pp , [7] N. Swangmuang and P. Krishnamurthy, Location Fingerprint Analyses Toward Efficient Indoor Positioning, Sixth Annual IEEE International Conference on Pervasive Computing and Communications, pp. 0 09, [8] M. B. Kjaergaard and C. V. Munk, Hyperbolic Location Fingerprinting- A Calibration-Free Solution for Handling Differences in Signal Strength, Sixth Annual IEEE International Conference on Pervasive Computing and Communications, pp. 0 6,