Presentation is loading. Please wait.

Presentation is loading. Please wait.

L OCATING IN F INGERPRINT S PACE : W IRELESS I NDOOR LOCALIZATION WITH L ITTLE H UMAN I NTERVENTION Zheng Yang, Chenshu Wu, and Yunhao Liu MobiCom 2012.

Similar presentations


Presentation on theme: "L OCATING IN F INGERPRINT S PACE : W IRELESS I NDOOR LOCALIZATION WITH L ITTLE H UMAN I NTERVENTION Zheng Yang, Chenshu Wu, and Yunhao Liu MobiCom 2012."— Presentation transcript:

1 L OCATING IN F INGERPRINT S PACE : W IRELESS I NDOOR LOCALIZATION WITH L ITTLE H UMAN I NTERVENTION Zheng Yang, Chenshu Wu, and Yunhao Liu MobiCom 2012 - Sowhat 2012.08.20

2 O UTLINE Introduction System Design Evaluation Discussion Conclusion

3 O UTLINE Introduction System Design Evaluation Discussion Conclusion

4 M OTIVATION RSSI fingerprinting-based localization Site survey Time-consuming Labor-intensive Vulnerable to environmental dynamics Inevitable Inevitable

5 O BJECTIVE Wireless Indoor Localization Approach RSSIFloor Plan User Movement

6 O UTLINE Introduction System Design Evaluation Discussion Conclusion

7 L I FS, S YSTEM A RCHITECTURE Geographical dist. ≠ Walking dist. RSSI + Distance

8 M ULTIDIMENSIONAL S CALING (MDS) Information visualization for exploring similarities/dissimilarities in data

9 S TRESS - FREE F LOOR P LAN MDS Geographical distance ≠ Walking distance, Ground-truth floor plan – conflict with measured distance Sample grids in a floor plan (grid length l = 2m) Distance matrix D = [d ij ], d ij = walking distance between point i and j Stress-free floor plan – 2D & 3D

10 F INGERPRINT S PACE – F INGERPRINT & D ISTANCE M EASUREMENT Fingerprints and distance collection Record while walking Footsteps every consecutive steps by accelerometer Set of fingerprints, F = {f i, i = 1~n} Distance(footsteps) matrix, D ’ =[d ’ ij ] Pre-processing Merge similar fingerprints (δ ij <ε) Accelerometer reading Twice integration  Distance: Noice Local variance threshold method  Step count Stride lengths vary?  MDS tolerate measurement errors

11 F INGERPRINT S PACE – F INGERPRINT S PACE C ONSTRUCTION Adequate fingerprints & distance 1. 10x sample locations in stress-free floor plan 2. First several days for training d ’ ij unavailable  d ’ ij = d ’ ik + d ’ kj Shortest path  update D ’ all-pairs of fingerprints Floyd-Warshall algorithm MDS  Fingerprint space 2D & 3D

12 M APPING – C ORRIDOR & R OOM R ECOGNITION F c Corridor recognition ( F c ) Higher prob. on a randomly chosen shortest path Minimum spanning tree Betweenness Watershed 1. Size(corridor) / Size(all) 2. Large gap of betweenness values F Ri Room recognition ( F Ri ) k-means algorithm (k = number of rooms) Classify fingerprints into the corridor or rooms

13 Fingerprints collected near “doors” P D = {p 1, p 2, …, p k }, stress-free floor plan F D, fingerprint space distance matrix D and D ’  l = ( l p1, l p2, …, l p k-1 ) l’ = ( l f1, l’ f2, …, l’ f k-1 ) cosine similarity M APPING – R EFERENCE P OINT Near-door fingerprints, F D, labeled with real locations 1.Map near-door fingerprints to real locations (F D → P D ) 2.Map rooms to rooms 1.Map near-door fingerprints to real locations (F D → P D ) 2.Map rooms to rooms

14 Floor-level transformation Stress-free floor plan ≠ Fingerprint space ∵ translation, rotation, reflection Transform matrix, x i = coordinate of f i ∈ F D y i = coordinate of p i ∈ P D For fingerprint with coordinate x real location = sample location closest to Ax + B Room-level transformation Room by room Doors and room corners as reference point Transformation matrix M APPING – S PACE T RANSFORMATION

15 O UTLINE Introduction System DesignEvaluation Discussion Conclusion

16 H ARDWARE AND E NVIRONMENT 2 Google Nexus S phones Typical office building covering 1600m 2 16 rooms, 5 large – 142m 2, 7 small, 4 inaccessible 26 Aps, 15 are with known location 2m x 2m grids, 292 sample locations

17 E XPERIMENT D ESIGN 5 hours with 4 volunteers Fingerprints recording – every 4~5 steps (2~3m) Accelerometer – work in different frequency based on detecting movement 600 user traces, with 16498 fingerprints Corridor, >500 paths Small rooms, >5 paths Large rooms, >10 paths Half of data used for training, half …………………... in operating phase

18 T HRESHOLD V ALUE OF F INGERPRINT D ISSIMILARITY

19 S TEP C OUNT 5 ~ 200 footsteps Error rate = 2% in number of detected steps Accumulative error of long path Unobvious performance drop ∵ only use inter-fingerprint step counts

20 F INGERPRINT S PACE 795 fingerprints when ε = 30

21 C ORRIDOR R ECOGNITION Refining Perform MST iteratively Sift low betweenness Until MST forms a single line

22 R OOM R ECOGNITION

23 R EFERENCE P OINT M APPING

24 P OINT M APPING 96 percentile < 4m Average mapping error = 1.33m 96 percentile < 4m Average mapping error = 1.33m

25 L OCALIZATION E RROR Emulate 8249 queries using real data on LiFS Location error Average, LiFS = 5.88m RADAR = 3.42m Percentile of LiFS 80 < 9m / 60 < 6m Caused by symmetric structure Fairly reasonable! Room error = 10.91%

26 O UTLINE Introduction System Design EvaluationDiscussion Conclusion

27 D ISCUSSION Global reference point Last reported GPS location Locations of APs Similar surrounding sound signature … Could be added in LiFS for more robust mapping Key for symmetric floor plans / multi-floor fuildings Large open environment

28 O UTLINE Introduction System Design Evaluation DiscussionConclusion

29 C ONCLUSION LiFS Spatial relation of RSSI fingerprints + Floor plan Low human cost Comments Clear architecture Not specific descriptions in evaluation

30 T HANKS FOR L ISTENING ~


Download ppt "L OCATING IN F INGERPRINT S PACE : W IRELESS I NDOOR LOCALIZATION WITH L ITTLE H UMAN I NTERVENTION Zheng Yang, Chenshu Wu, and Yunhao Liu MobiCom 2012."

Similar presentations


Ads by Google