Indoor Location Estimation Using Multiple Wireless Technologies 17/01/2019 Indoor Location Estimation Using Multiple Wireless Technologies Dhruv Pandya, Imperial College {dhruv.pandya@doc.ic.ac.uk} Ravi Jain, Telcordia Technologies {rjain@acm.org} Emil Lupu, Imperial College {e.c.lupu@doc.ic.ac.uk}
Location-aware Computing Proliferation of mobile devices + development of standard low-cost wireless networks e.g. Bluetooth Richer application set User customization Example applications: E911: locating callers in emergencies, Call forwarding, Find nearest restaurant Single-technology Approaches Infrared transmissions: Active Badges, Want 92 Time of Flight of Ultrasound: Cricket, Priyantha 2000 Time of Flight of RF, Triangulation: GPS RF, Scene Analysis: RADAR 17/01/2019 Dhruv Pandya
17/01/2019 Current Limitations Technology limitations e.g. GPS does not work indoors Application restricted to sensor technology and its accuracy System-dependent location representation 17/01/2019 Dhruv Pandya
Multiple Technologies Devices with multiple wireless technologies Application flexibility Improved location accuracy Location accuracy Jain ICDE 2001 Pandya June 2002, Masters’ Thesis Similar approaches: Geometric modelling: Leonhardt 1998 Conflict resolution of multiple sources: Myllymaki 2002 17/01/2019 Dhruv Pandya
Static Scene Analysis – Data Collection 17/01/2019 25.7m Offline Samples 49 physically distinct locations 50 signal samples from each base station, in each of four directions Run-time Samples 6 locations per three, two and one base station coverage scenario 50 samples from each base station while facing the North direction Corridor 32.2m As used in RADAR. The disadvantage of this approach is that certain areas may receive more offline locations. Hence those areas may be favoured more. Make a note about how bluetooth points are used as access points. Runtime samples: we account for temporal and spatial variability of wireless links. Signal Characteristics 802.11: signal strength in dBm Bluetooth: ‘link quality’ Offline location 802.11 Base Stations Run-time location Bluetooth Base Stations 17/01/2019 Dhruv Pandya
Final location estimate Smallest Polygon 17/01/2019 Runtime A Comparison Function Corridor Offline Final location estimate C B 17/01/2019 Dhruv Pandya
Triangulation Runtime A Comparison Function Corridor Offline C B 17/01/2019 Dhruv Pandya
Final location estimate Nearest Neighbor Runtime A Comparison Function Corridor Offline Final location estimate C B 17/01/2019 Dhruv Pandya
Basic Fusion Comparison Resolve estimates ‘Fuse’ 17/01/2019 Dhruv Pandya
Performance Evaluation Distance Error Average errors for locations with similar coverage scenario 17/01/2019 Dhruv Pandya
Comparison of Algorithms Reduction in error with increased base stations SP has best accuracy under three (1.8-2m) and one (4-5m) base station coverage 17/01/2019 Dhruv Pandya
802.11 Vs Bluetooth Bluetooth performs better than 802.11 for all algorithms, except SP 17/01/2019 Dhruv Pandya
Basic Fusion Basic Fusion performs better than TN (1.7-3.3m for three, 0.2-2.5m for two) and NN (0.5-2m for three, 0.4-2.3m for two) 17/01/2019 Dhruv Pandya
Fusion Vs Single-technologies Bluetooth signal characteristic: Link Quality 17/01/2019 Dhruv Pandya
Summary Location sensor technology restricts applications: Use multiple technologies Experimental investigation into improving location accuracy by data fusion using Bluetooth and 802.11 wireless LAN Differences to RADAR Introduce spatial and temporal variability in runtime samples Attempt to represent a more realistic situation Single Technology Smallest Polygon works best with 802.11 wireless LAN Bluetooth works well with Triangulation and Nearest Neighbor Multiple Technologies: Increased base stations: accuracy increases 0.2-3.3m Fusion improves accuracy over 802.11 (by 0.4-1.5m), but looses out over Bluetooth 17/01/2019 Dhruv Pandya
Further Work Sophisticated single-technology and fusion algorithms Improved location accuracy metrics User A is 50% likely to be 2m radius of an estimate Confidence Precision 17/01/2019 Dhruv Pandya
Acknowledgement Dhruv Pandya acknowledges travel funding support from EPSRC project GR/R95715/01 AEDUS: Adaptable Environments for Distributed Ubiquitous Systems 17/01/2019 Dhruv Pandya
Questions dhruv.pandya@doc.ic.ac.uk 17/01/2019 Dhruv Pandya
Single-technology Approaches Infrared transmissions: Active Badges Time of Flight analysis of Ultrasound: Cricket Triangulation, RF: GPS Scene Analysis, RF: RADAR 17/01/2019 Dhruv Pandya
Presentation Outline Motivation Related work Data collection 17/01/2019 Presentation Outline Motivation Related work Data collection Location Estimation and Fusion algorithms Performance Evaluation Conclusion 17/01/2019 Dhruv Pandya
Static Scene Analysis – Bracketing 17/01/2019 Static Scene Analysis – Bracketing Location Base Station ID x y SS where r = Average of 50 Run-time SS samples, b = bracket value determined by the standard deviation of SS in our environment Select candidate(s) that match the first SS in {r-b,…,r,…r+b}. The process may return more than one location because it is possible that more than one location may match the first SS. the location(s) that match the first possible SS in Q 17/01/2019 Dhruv Pandya
Smallest Polygon 17/01/2019 Dhruv Pandya 17/01/2019 SP works under the following theory: if there was no interference then for a given location, candidates from all BS would be the same. SmallestPolygon is a function that takes as input one or more sets of location estimates. Thus we want to find the candidates that are closest to each other. Base case: if Ea is the only non-empty set, then selects one candidate randomly. 17/01/2019 Dhruv Pandya
Triangulation 17/01/2019 Dhruv Pandya
Nearest Neighbour 17/01/2019 Dhruv Pandya 17/01/2019 Note that RADAR consider offline locations. We consider runtime locations, temporal and spatial variability. Radio interference at one point in time will not be the same as another – that’s an assumption. 17/01/2019 Dhruv Pandya
Standard Deviation of Algorithms 17/01/2019 Dhruv Pandya