INDOOR LOCALIZATION OUTLINE

Transcription

1 INDOOR LOCALIZATION DHARIN PATEL VARIL PATEL OUTLINE INTRODUCTION CHALLAGES OF INDOOR LOCALIZATION LOCATION DETECTION TECHNIQUE INDOOR POSITIONING ALGORITHM RESEARCH METHODOLOGY WIFI-BASED INDOOR LOCALIZATION SYSTEM IMPLEMENTATION IN JAVA CALCULATION OF ESTIMATION ERROR ALGORITHM TO REDUCE ERROR COMPARISON AND CONCLUSION 1

2 INTRODUCTION COMPONENTS Communication Network Mobile Device (receiver) Positioning Device CHALLAGES OF INDOOR LOCALIZATION WALLS MOVEMENTS OF HUMANS DOOR FURNITURE EQUIPMENTS OTHER FACTORS 2

3 LOCATION DETECTION TECHNIQUE GPS Laser scanning Laser scanners are used on particular object to find the location information objects, laser scanning takes high resource utilization like time, CPU and memory Radio frequency identification and detection INDOOR POSITIONING ALGORITHM Triangulation Proximity Fingerprint based indoor localization. 3

4 TRIANGULATION The triangulation algorithm estimate the location of the target place based on geometric properties of triangles. When the mobile device at the target place receives the Wi-Fi signals from one or more Wi-Fi access points then the target place can be estimated by triangulation Proximity Proximity Algorithm: When mobile device receive Wi-Fi signal from various Wi-Fi access points at target location, at that time one Wi-Fi access point gives strongest signal strength so location of that access point is considered as location of object. In short this method use proximity between Wi-Fi access points and target place. 4

5 Fingerprint Based Indoor Localization The two main function of Fingerprinting Based Indoor Localization. to collect finger print from the surrounding location matched with original database There are two phase in the Fingerprinting Based Indoor Localization. (1)offline (2)online. RESEARCH METHODOLOGY Place: a HKUST Academic Indoor building Area: 145.5m 37.5m. Number of Grids: 247. Size of a Grid: 1.5m 1.5m Training dataset: 2322 fingerprints part of offline phase Test dataset:53 fingerprints part of online phase 5

6 Fingerprint Based Indoor Localization There are two process Site survey (Fingerprints are measured and recorded at every location and finger print database is accordingly constructed) Finger print matching Fingerprint Based Indoor Localization User sends current RSS fingerprint and comparing it using localization algorithm with fingerprint database and it returns the location of matched fingerprint of the user s position. A fingerprint f={fi(i=1,2.n)} (fi is the RSS value of access points which appears in fingerprints) 6

7 Data representation: :-92 67:-85 71:-60 73:-87 75:-74 76:-91 77:-94 79:-91 (Location Label)= 191 (AP s Index_1)= 61 (RSS value)= -92 (AP s Index_2)=67 (RSS value)=-85 Wi-Fi based Indoor Localization System f = fingerprint from current location f= database fingerprint RSS difference di= fi-fi d =sqrt{(ss1-ss1 )^2+(ss2-ss2 )^2+(ss3-ss3 )^2 }. After matching all these fingerprints the fingerprint f that achieve the highest match with the query fingerprint f. 7

8 Wi-Fi based Indoor Localization System The user s location is estimated as the corresponding location L(f*) of f* and by assuming we can get true location of f is L(f). Location estimation error is given by e= L(f)- L(f*). JAVA Programming Training dataset=2322 Test dataset=53 Manually it is difficult to compare but using java program we have compared each fingerprints of test dataset with training dataset. After comparison using RSS value we found most closest fingerprints from training dataset which match with test dataset. 8

9 JAVA Code JAVA Code 9

10 JAVA Code JAVA Code(comparing string) 10

11 OUTPUT OF PROGRAM CALCULATION we took particular test line to do calculation and it is :-93 23:-87 62:-9166:-86 67:-93 Our output gives appropriate RSS difference with training lines and least value is refers to following closest training line :-96 23:-83 45:-87 48:-87 62:-91 66:-84 67:-93 11

12 ALGORITHM TO REDUCE ERROR In normal calculation we just consider one fingerprint and compare with test dataset fingerprint, but in nearest neighbor method we consider those fingerprints whose RSS closely match with each test dataset. There are many fingerprints available, here we take first five closest fingerprints. Estimation location is the fingerprint which has the least error distance among all fingerprints in the radio map. 12

13 ALGORITHM TO REDUCE ERROR There are many points available which are at same distance from target location and here in each fingerprint we get variation in signal strength value so we cannot pick only single point from all because all are close to each other, location of every point is different so if we take average of co-ordinate then we can bring point more close to the true location estimated location = average of loc(f1)+ loc(f2) + loc(f3) ALGORITHM TO REDUCE ERROR In same way we have 53 fingerprints so for each finger print we found new locations using average of first three and five location and also find estimation error distance for new location We also found estimation error distance between location of actual fingerprint and closest finger print from training dataset and we compare all three result with each other. 13

14 ERROR DISTANCE(METERS) 12/19/2016 COMPARISON ERROR FOR 3 LOCATION AVERAGE(METERS)" ERROR FOR 5 LOCATION AVERAGE(METERS) ERROR DISTANCE(METERS) FINGERPRINTS Result table Average of first 3 location Average of first 5 location Normal Value from closest fingerprints Maximum error(m) Minimum error(m) StdDev Median Average Error(m)

15 CONCLUSION Comparison and calculation of Euclidian distance we got 53 different value of error distance by averaging 3 and 5 closest locations using nearest neighbor method. Comparing all outputs with each other we can conclude that nearest neighbor method gives more accuracy while averaging three location instead of five location. REFERENCES [1] RADAR: An in building RF based user location and tracking system Paramvir Bahl and Venkata N. Padmanabhan, Microsoft Research, IEEE Infocom, [2] Recent Advances in Wireless Indoor Localization Techniques and System by Zahid Farid, Rosdiadee Nordin, and Mahamod Ismail [3] An Improved WiFi Indoor Positioning Algorithm by Weighted Fusion Rui Ma *, Qiang Guo, Changzhen Hu and Jingfeng Xue [4] DorFin: WiFi Fingerprint-based Localization Revisited Chenshu Wu_, Zheng Yang_, Zimu Zhouy, Liu_ and Mingyan Liu [5] Yunhao 15

16 16