Indoor Positioning by LED Visible Light Communication and Image Sensors

Transcription

1 International Journal of Electrical an Computer Engineering IJECE Vol., No., December 0, pp. 6~70 ISSN: Inoor Positioning b LED Visible Light Communication an Image Sensors Mohamma Shaifur Rahman, M. Mejbaul Haque, Ki-Doo Kim Dept. of Electrical an Electronic Engineering, Khulna Universit of Engineering & Technolog, Khulna-90, Banglaesh Dept. of Electronics Engineering, Kookmin Universit, Seoul 6-70, Korea msrbitk@ahoo.com, mhmukul.eee05@gmail.com, kk@kookmin.ac.kr Abstract High power white LEDs are epecte to replace the eisting lighting technologies in near future which are also suggeste for visible light communication VLC. We propose an algorithm for high precision inoor positioning using lighting LEDs, VLC an image sensors. In the propose algorithm, four LEDs transmitte their three-imensional coorinate information which were receive an emoulate b two image sensors near the unknown position. The unknown position was then calculate from the geometrical relations of the LED images create on the image sensors. We escribe the algorithm in etails. Simulation of the propose algorithm was one an presente in this paper. This technique i not require an angular measurement which was neee in contemporar positioning algorithms using LED an image sensor. Simulation results showe that the propose sstem coul estimate the unknown position within the accurac of few centimeters. Positioning accurac coul be increase b using high resolution image sensors or b increasing the separation between the image sensors. Kewors: Inoor positioning, LED, Image sensor, VLC.. Introuction Inoor positioning sstems are becoming increasingl significant for pervasive an contet-aware computing with location awareness, autonomous robot movement, sensor applications etc. During the last two ecaes, researchers have been tring to evelop positioning techniques to provie precise location information using GPS, IR, RFID, Blutetooth, WLAN, Ultrasoun, an so on []. GPS has been etensivel use for positioning in outoor environment such as in car navigation, mobile phones, ships, planes, surveing of public works etc. However, since GPS positioning epens on raio wave propagation, multipath faing an isturbances from other raio sources lea to large positioning error in inoor environment. The conventional inoor positioning sstems such as WLAN, RFID, Bluetooth an Ultrasoun sstem have problems ue to sstem instabilit, long response time, low accurac an low precision. Recentl, visible light communication VLC is emerging as an attractive technolog for inoor positioning sstem [-6]. Together with illumination, lighting equipments serve the purpose of ata communication in VLC sstem. This sstem is regare as one of the most promising alternatives to ban limite raio wave communication since it oes not cause or suffer from raio or electromagnetic interference. VLC can be use in builings, unergroun an even in hospitals where raio frequenc is prohibite. Compare to conventional lighting evices such as incanescent an fluorescent lamps, LED is more avantageous in terms of long life epectanc, high tolerance to environmental haars, low operating voltage, low power consumption an nominal heat issipation lighting. Hence, high power white LEDs are consiere to be a strong caniate for the future illumination technolog [7-9]. Therefore, we focus on eveloping VLC base positioning sstem. In this paper, we propose an algorithm for high precision inoor positioning using LED base VLC an image sensors. The propose scheme utilies LED lighting arra as VLC transmitter that transmits the three-imensional D coorinate information of some reference LEDs. Two image sensors receive the information from all the reference LED lights through lenses an emoulate the position information. The require unknown position is then calculate using the position information of the reference LEDs an the geometric relationship of the images on the two image sensors.. Previous Works. VLC base positioning research is still at a preliminar stage. Rigorous works are et to be publishe. However, some significant VLC base positioning techniques an algorithms can be foun in [-6]. [] an [] escribe the VLC base positioning sstem using visible light communication ientification VLID, photo etector an 6-aes sensor. These works escribe that three ifferent estimate receiver positions ERP are create ue to the effect of the receiver s fiel-of-view FOV. In the conventional positioning scheme, a single ERP is use where the onl wa to reuce the estimation error is b ecreasing the FOV of the receiver []. The authors in [] propose switching estimate receiver position scheme where the ERP is switche epening on the receiver s tilt angle, to limit the estimate error istance. Root mean square of error istances were foun to be 6. cm,.6 cm an 9.8 cm for receiver FOV angles of 5º, 7.5º an 0º, respectivel. The scheme estimates two-imensional D positions. However, the accurac of the estimate positions is unavoiabl epenent on accurac of Receive Sep 8 th, 0; Revise Nov 0 th, 0; Accepte Nov th, 0

2 6 ISSN: angulations ata from 6-aes sensor. A VLC guiance sstem using fluorescent lamps an photo sensor has been propose in [], which can also estimate the D location with accurac between 0 to 0 cm ranges. The sstem is esigne to assist the visuall impaire persons b proviing them the necessar information insie the builing. Location estimation in this sstem requires at least three fluorescent lamps to be etecte b the photo sensor which implies some limitations on positioning everwhere insie the room ue to the geometr of the lighting installation. Positioning accurac of this sstem obviousl epens on the accurac of measuring the tilt angle of the photo sensor. VLC base positioning using LED lights an image sensor has been propose in [5] an [6]. This sstem performs the optical intensit moulation of the LED illumination for ata transmission an receives the light b using image sensor. Data from various light sources can be receive at the same time b using image sensor as receiver. This is one b spatiall separating the light ras from ifferent LEDs b a lens provie that the receiver is irecte towars the light source. In [5], at least three LEDs from the LED lighting arra transmit the D coorinate information. The D image sensor receives the lights separate spatiall b a lens an emoulates the D coorinate information of each LED. The position of the receiver is then calculate b solving two sets of quaratic equations which ma leave some possibilities of computational error. In [6], an aitional 6-aes sensor is use to etermine the receiver s irection along with its position. This scheme uses collinearit conition to relate the D coorinates of the LEDs to the D coorinates of the image sensor. Simulation results show that the receiver s position can be estimate with accurac of within.5 meters if the piel sie is 6 0 m. Analsis of the previous works reveals that positioning error of the previousl propose schemes ma occur from the precise angular measurements an solution of quaratic equations for position estimation. Hence, we propose a new algorithm for positioning in VLC sstem base on lighting LEDs an two image sensors which oes not require an angular measurement an can overcome the limitations of the eisting schemes. 6. Propose Algorithm In the propose sstem, LED arras use for illumination are also use for positioning. At least four LEDs from the arra transmit their D coorinate information which are receive an emoulate b two image sensors through two optical lenses. In choosing the LEDs, we have to make sure that the are not collinear. Let us consier that A, B, C an D in Fig. represent four LEDs which transmit their known D coorinates,,,,,,,, an,, respectivel. A,, LED arra B,, C,, D,, Unknown position Lens p U,, p Image sensor Image sensor Figure. Propose VLC positioning sstem The image sensors use for receiving the light signals consist of a D arra of photosensitive elements. Hence, each element or piel can act as an iniviual photo sensor an multiple LED signals can be etecte an emoulate simultaneousl b the single image sensor. Two image sensors receive the intensit moulate lights from the four LEDs separate spatiall b using two separate lenses an emoulate the D coorinate information of the LEDs. The image sensors are installe in the same plane at a known lateral istance with their major ais in the same line p p. Lens, having ientical properties an focal length, is installe above the center of each image sensor. The ais IJECE Vol., No., December 0 : 6 70

3 IJECE ISSN: of each lens, normal to the image sensor plane, intersects the center of the corresponing image sensor. The phsical arrangement of the sstem can be unerstoo from Fig.. U is the mi point of the straight line joining the centers of the two lenses. The unknown D coorinate of the point U,, is estimate in this scheme. The istances of the point U from the four reference LEDs A, B, C an D are enote b,, an, respectivel. These istances are calculate from the geometric relationship between the istance an the position ifference of the LED images on the two image sensors. Geometric relationship of the propose positioning scheme is etaile in Fig.. This figure eplains the proceure for etermining the istance of the point U from an LED. Here, the focal length of the lenses, f an the istance between the centers of the lenses, L are known. h a b Projection of i f θ c L f Projection of i i i Image sensor Image sensor Figure. Positioning Technique If the istances of the image centers from the center of the image sensors are i an i, an their projection on the major ais of the image sensors are Pian Pi, the istance can be calculate as follows. i Pi Pi f L h i f i c f i a b h c f h f 5 6 θ cos a L b al a L L a cosθ 7 8 Inoor Positioning b LED Visible Light Communication an Image Sensors Mohamma Shaifur Rahman

4 ISSN: IJECE Vol., No., December 0 : Similarl,, an can be obtaine from the other three LEDs. Hence, we obtain a set of four quaratic equations as follows: 9 0. Position Estimation. Least Square Estimation To estimate the D unknown positions, a set of three equations are sufficient if the istances coul be estimate accuratel. However, ue to quantiation error in the piels, eact istance between the LED an the unknown position can not be estimate. The intuitive solution to this problem is to use more than three references. Mathematicall, this turns the above equations into an over-etermine sstem of equations, terme as linear least square LS problem. We can estimate the unknown position b solving the following equation for X. D M MX M T T where, M, D an X Conition for linear LS solution is that matri M shoul have full rank. However, since the reference LEDs are in the same plane ie. all have same coorinate value, this conition oes not satisf. Hence, we a a small value with the ais value of the fourth reference LED. Even though this technique solves the singularit problem of matri M, we obtain an erroneous estimation for ais. To estimate the ais value, we further use vector estimation. Three reference LEDs, A, B an C are use for this estimation.. Vector Estimation If A, B an C represent the vectors that correspon to the coorinates of A, B an C, respectivel, two intersecting points can be calculate using the following equations to. A B BA A C CA 5, CA BA D cross 6 I i D E 7 X *CA i BA i 8 Y *BA i i CA 9 E cross D Y X U, 0

5 IJECE ISSN: U i * D i V E P A U V P A U V Vectors P an P represent the two intersecting points from which ais value of one is selecte epening on the geometr of the room. 5. Simulations an Results 5.. Simulation Setup Performance of the propose algorithm is evaluate b simulation using Matlab. One important issue of this scheme is the area within which the position can be etermine uner an LED arra. If the four reference LEDs are seen b both the image sensors, the sstem can make the estimation. The positioning area is reuce to the minimum when the image sensors are parallel to the LED arra. Figure escribes the geometric imensions for calculating the working area. p q LED arra Sie View ϕ r h Top View Figure. Positioning area uner an LED arra Assuming that the image sensors are parallel to the LED arra, we can calculate the positioning area from the following equations. p h*tanϕ r p p q 5 where, ϕ is the half of FOV of the image sensors, q is the length of a sie of the LED arra, h is the vertical istance between the LED arra an image sensors, r is the length of a sie of the positioning area. In our simulation moel, an LED arra of m area is assume. We consier the FOV of the image sensors to be 5º an the vertical istance between the LED arra an the image sensor to be.5 m. Therefore, the propose scheme will be able to estimate position within an area of approimatel.8. 8 m,.5 m below the LED arra. Hence, keeping istance between two ajacent positions to cm on both the horiontal aes, we assume eperimental positions on the horiontal plane. In our simulation, the major ais of the image sensors is assume to be parallel to the -ais of the positioning area. We choose APS-C tpe 6 Mega piel image sensor, which is use commerciall in Nikon D0, D50, D70, D70s, Penta K00D, etc. The simulation parameters are liste in Table. The piels on the image sensor are arrange in D arras. LED images can be constructe b multiple piels. For this reason, we nee to etermine the center of each image prior to other measurements. Let, N piels constitute the image of a particular LED. The center of the image is then calculate in terms of row an column number of image matri b calculating the centroi. Inoor Positioning b LED Visible Light Communication an Image Sensors Mohamma Shaifur Rahman

6 66 ISSN: Parameter Table. Simulation Parameters Value Image sensor imension mm Number of piels FOV 5º Focal length of lens Distance between center of two lenses Vertical istance of image sensor an LED arra.5 cm 0 cm.5 m Positioning area.8. 8 Distance between ajacent positions in -ais Distance between ajacent positions in -ais m cm cm Total estimating positions 76 N N r, N c N rj, N cj, j,,ll, N N j 6 where, N, N is the center of the image in terms of row an column number of the image matri, N, N are r c rj cj the row-column number of the piels that construct the corresponing LED image, an j is the number of piels. 5.. Least Square Estimation Results The performance of the propose algorithm with least square estimation is first evaluate for a fie unknown position. Keeping the other factors constant, the number of piels per line of the image sensor is increase from 500 to 008 with a step of 0 piels. Simulation result is shown in Fig Error along ais Error along ais Error along ais Positioning error Piels in one line Figure. Positioning errors using least square estimation for a fie unknown position. Error is fluctuating ue to quantiation error from the piels of the image sensors. Error along ais is higher than that of an aes IJECE Vol., No., December 0 : 6 70

7 IJECE ISSN: Horiontal ais represents piels per line an vertical ais represents the positioning error in meter. It is observe that the positioning error is ecrease with increasing the piel number. The error has a fluctuating behavior which is occurre ue to the quantiation error of the piel. The reason is that the image center is assume to be at the center of the piel which is obtaine from image center estimation using Eq. 6. However, the actual center of the LED image is not alwas at the piel center. As a result, we get the quantiation error. From the Fig., it is seen that the estimation error is 0.56, 07 an m along the, an aes, respectivel, when piels per line is 98. Overall error at this case is 0.88 m. Errors along an aes are much lower than that along ais. To avoi the singularit problem in the least square estimation iscusse in section., we ae a small length to the coorinate of the reference LED, D. For that reason, we obtain large error along ais. As a result, estimation error is increase in LS estimation. To solve that problem, we further estimate the unknown ais using the vector estimation an the ata from the three reference LEDs A, B an C. 5.. Vector Estimation Results For vector estimation of the D unknown position, we nee onl three reference LEDs. Hence, ata corresponing to reference LEDs A, B an C are use. As before, the piel per line is increase from 500 to 008 to unerstan the effect of piel sie on estimation error. All other parameters are kept unchange as LS estimation. Results are shown in Fig. 5. Horiontal an vertical aes inicate number of piels per line an estimation error, respectivel. The errors along, an aes are foun to be 0., 78 an m, respectivel, when the piel number is 98 per line. The overall error is 0.6 m. Also, it is seen in Fig. 5 that the vector estimation gives lower error along ais than that of LS estimation. On the other han, error along an aes are higher than that of LS estimation. In our simulation, we assume the major ais of the image sensor parallel to the ais. Due to this reason, error along ais is higher than that along ais. Summariing these two sets of results, we suggest the use of LS estimation for an aes, an vector estimation for ais that can give us a better overall estimation of unknown position. Estimation error from the combination of LS an vector estimation along the ifferent aes an the overall estimation error are shown in Fig. 6. Overall estimation error for the particular position is 0.56 m as the number of piels per line is E rror along ais E rror along ais E rror along ais P ositioning error P iels in one line Figure 5. Positioning error along the three ifferent aes for a fie unknown position.the positioning is one using vector estimation. Error along ais is much lower than that of least square estimation while error along an aes are higher 5.. Relation Between Lens Distance an Positioning Error Distance between the centers of the lenses certainl has some effect on positioning error. To fin out the relationship between lens istance an position error, we var the istance between the centers of the lenses from Inoor Positioning b LED Visible Light Communication an Image Sensors Mohamma Shaifur Rahman

8 68 ISSN: cm to 0 cm with step of cm an calculate the root mean square of error istance E rms of the 76 eperimental positions escribe in section., using the following equation. E n rms Err n n i E rror along ais E rror along ais E rror along ais P ositioning error Piels in one line Figure 6. Positioning error obtaine b using least square an vector estimation together, first one is for an aes value estimation an the later one is for ais value estimation RMS error, E rms [m] a RMS error for vector est. RMS error for LS-vector est. RMS error, E rms [m] b RMS error for LS-vector est. Eponential fitting Distance between lens centers, L [m] Figure 7. a RMS positioning error for vector an LS-vector estimation at ifferent lens istances b Secon orer eponential eca fitting of RMS error for LS-vector estimation IJECE Vol., No., December 0 : 6 70

9 IJECE ISSN: where, n is the number of position to be estimate an Err n is the estimation error of n th position. Results shown in Fig. 7a inicate that the RMS positioning error ecreases both for vector estimation an LS-vector estimation as we increase the istance between the two lenses. The improvement of the positioning performance is significant up to the istance of approimatel 0 cm. At 0 cm istance, the RMS errors for vector an LS-vector estimation are foun 0.9 an 9, respectivel. It is also foun that the RMS error of LS-vector estimation is lower than that of vector estimation if the lens istance is increase beon 7. cm. Since there is a trae-off between the positioning area an istance of the image sensors, so, we can keep the istance as low as possible. To establish a relationship between the RMS error an lens istance, we further performe secon orer eponential eca fitting on the LS-vector estimation RMS error results as shown in Fig. 7b. The relation is given below. E rms L / 959 L / e.59e 8 6. Discussion In this section, we have iscusse about some limitations of the propose algorithm an compare it with other previousl propose positioning algorithms base on VLC. 6.. Limitations Distance estimation in the propose algorithm epens on the image istance measurements on the image sensors. For the accurac of the estimation, each component nees to be installe with high measurements accurac. Image sensors shoul be installe in the same plane an the lenses shoul have ientical properties. Otherwise the positioning accurac woul be egrae largel. One major limitation of the VLC base positioning is the positioning area of the sstem. Since the positioning process solel epens on the reference coorinate information from the lighting source, the propose scheme can not calculate the position if all the reference LEDs can not be seen b both the image sensors. The lighting arrangement bears much importance for positioning since there shoul not be an shaowe region insie the positioning area. Another important limitation originates from the quantiation error of the piels. The LED ras can be projecte to an part of the piel of an image sensor; it is not alwas projecte to the center of the piel. Since the esire position is calculate assuming that the center of the LED image is in the center of a piel, it causes the quantiation error. The propose algorithm is applicable for stationar or slowl moving targets. To appl this sstem for moving targets, ata transmission rate of the VLC sstem shoul be sufficientl high. 6.. Comparison with Other Technologies In this section, we follow-up an iscuss about the avantages of our propose algorithm. Some previous VLC base positioning algorithms [-6] are briefl iscusse in section. We trie to solve some major challenges b our propose algorithm. In [] an [], photo etector is use as receiver which can receive signal from a single light source at a time. In the propose algorithm, use of image sensor facilitates to receive signals from several light sources simultaneousl. In case of positioning error, switching estimate receiver position scheme [] reports the D RMS positioning error to be 6. cm,.6 cm an 9.8 cm for receiver FOV angles of 5º, 7.5º an 0º, respectivel. The VLC guiance sstem using fluorescent lights an photo sensor [] estimates the location with accurac within 0 to 0cm. However, the positioning is D. Positioning algorithm using LEDs, image sensor an 6-aes sensor [6] estimates the receiver s position with accurac of within.5 m provie that piel sie is m. In our propose scheme, D positions can be estimate with an RMS error of about few centimeters. In aition, positioning accurac of the sstems escribe in [,,, 6] obviousl epens on the accurac of angular measurements. Table. Comparison of the propose algorithm with earlier VLC base positioning algorithms Angular Ref No. Transmitter Receiver Positioning tpe Accurac measurement LED Photo etector D ~m Neee LED Photo etector D 0.98~0.6m Neee Fluorescent lamp Photo etector D 0.~0.m Neee 5 LED Image sensor D Not neee 6 LED Image sensor D.5m Neee Propose algorithm LED Image sensor D 0~0.5m Not neee Our propose algorithm oes not nee an kin of angular measurement. Rather, the position is calculate from the ifference of the image istances in two image sensors. Hence, the measurement error is significantl reuce since we o not nee an angular measurement. Some major features of the propose algorithm are compare with the Inoor Positioning b LED Visible Light Communication an Image Sensors Mohamma Shaifur Rahman

10 70 ISSN: previous VLC base positioning sstems an are liste in Table. Comparison of the properties liste in Table epicts that the propose sstem woul be avantageous ue to high accurac an angular measurement-less D positioning. 7. Conclusion In this paper, we propose an algorithm for inoor positioning using lighting LED arra an image sensors. White LEDs are use as optical transmitters in which at least four spatiall separate non-collinear LEDs sen their known position information. The image sensors locate at the unknown position receive an emoulate the light signal through lenses. The unknown position is then calculate using the geometric relations of the LED image istances. Mathematical formulations for the propose algorithm are iscusse in etails. A series of simulations are eplaine to unerstan the positioning error characteristics. Simulation results inicate that b using the propose algorithm, the positioning error can be minimie within the range of a few centimeters. The propose technique has the benefits of evice simplicit. References [] J. Hightower, an G. Borriello, Location Sstems for Ubiquitous Computing, Computer, vol., no. 8, pp.57-66, 00. [] C. Sertthin, E. Tsuji, S. Haruama, an M. Nakagawa, Visible Light Communication an 6-aes Sensor Base Positioning Technique for a Barcoe Reaer Sstem, IEICE Tech. Report, vol. 07, no. 50, pp. 9-, 008. [] C. Sertthin, E. Tsuji, M. Nakagawa, S. Kuwano, an K. Watanabe, A Switching Estimate Receiver Position Scheme for Visible Light Base Inoor Positioning Sstem, Proc. th Int. Conf. Wireless Pervasive Computing, Melbourne, Australia, Feb. 009, pp [] X. Liu, H. Makino, an Y. Maea, Basic Stu on Inoor Location Estimation Using Visible Light Communication Platform, Proc. 0th Annual Int. IEEE MEBS Conf., Vancouver, British Columbia, Canaa, Aug. 008, pp [5] S. Horikawa,T. Komine, S. Haruama, an M. Nakagawa, Pervasive Visible Light Positioning Sstem Using White LED Lighting, IEICE Tech. Report, vol. 0, no. 7, pp. 9-99, 00. [6] M. Yoshino,S. Haruama, an M. Nakagawa, High-Accurac Positioning Sstem Using Visible Le Lights an Image Sensor, IEEE Raio an Wireless Smposium, Orlano, FL, USA, Jan. 008, pp. 9-. [7] M. Ishia, InGaN Base LEDs an Their Application, Optronics, vol. 9, no. 8, pp. 0-5, 000. [8] T. Taguchi, Technological Innovation of High-Brightness Light Emitting Dioes Les an A View Of White LED Lighting Sstem, Optronics, vol. 9, no. 8, pp. -9, 000. [9] T. Tamura, T. Setomoto, an T. Taguchi, Funamental Characteristics of The Illuminating Light Source Using White LED Base on Ingan Semiconuctors, Trans. IEE Japan, vol. 0-A, no., pp. -9, 000. Bibliograph of authors Mohamma Shaifur Rahman receive his B.Sc. an M.Sc. Engineering egree in 00 an 005, from Khulna Universit of Engineering & Technolog KUET, Banglaesh. He joine as a Lecturer in the Department of Electrical an Electronic Engineering of the same Universit in 00. He got the PhD egree in 00 from Kookmin Universit, Republic of Korea. He is now working as an Associate Professor in KUET. His current research area inclues position estimation, igital signal processing, G communication technologies, wireless sensor network, etc. He is a life member of Institution of Engineers, Banglaesh. M. Mejbaul Haque receive his B.Sc. Engineering Degree from Rajshahi Universit of Engineering & Technolog RUET, Banglaesh, in April 9, 00 in Electrical an Electronic Engineering. He publishe several papers. He is currentl working as a Lecturer in Electrical an Electronic Engineering Department, KUET. His research interests inclue MIMO-OFDM, visible light communication, power sstem stabilit, renewable energ etc. He is an associate member of Institution of Engineers, Banglaesh. Professor Ki-Doo Kim receive his B.S. egree in Electronics Engineering in 980 from Sogang Universit, Republic of Korea. He then worke as a research engineer at the Agenc for Defense Development in Republic of Korea from 980 to 985. He got the MS an Ph.D egree in 988 an 990 from Pennslvania state universit, USA. He then joine Kookmin Universit in 99. He also serve as a visiting scholar in the Department of Electrical an Computer Engineering, Universit of California, San Diego, USA from 997 to 998. He is now working as a Professor in the School of Electrical Engineering of Kookmin Universit, Republic of Korea. His current research interest inclues igital communication, igital signal processing, etc. IJECE Vol., No., December 0 : 6 70