With the emergence of new standards for mobile sensors and their increased availability, until now unexplored opportunities of generating high quality Indoor Localization data are coming within reach.
Learn how crowd-sourced big data merged with building information data can be utilized to predict human actions in indoor environments.
Bajaj Allianz Life Insurance Company - Insurer Innovation Award 2024
Big Data for crowd mobility
1. indoor positioning and navigation for mobile apps
indoo.rs GmbH, Austria
Thomas Burgess
<thomas@indoo.rs>
7th GeoIT Wherecamp Conference 2017
Chief Research Officer
Big data for indoor
crowd mobility
2. Outline.
üindoo.rs
➡ Who we are, what we do, who our customers are
üLocalization
➡ How the blue dot is calculated
üRadio maps
➡ Crowdsourcing radio data
üCrowd mobility
➡ Intelligence from crowd data
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th 2
3. indoo.rs
Who we are, what we do, who our customers are.
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th
4. üThomas Burgess
➡ PhD in particle physics
➡ Chief Research Officer
➡ At indoo.rs since 2013
➡ Swede living in Austria
üindoo.rs GmbH
➡ Technology startup since 2010
➡ ~20 staff / ~5 researchers
➡ Based in Vienna, Austria
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th 4
Who is talking? Me!
5. What we do?
üEnabling location awareness
➡ iOS/Android mobile SDK
➡ Main data source iBeacons (or WiFi)
üNavigation
➡ Interactive for use in foreground applications
➡ Accurate for human tracking
➡ Real time on device (works offline)
üAsset tracking
➡ Active infrastructure, or indirect via Navigation
➡ Server side calculations on demand
➡ Track anything with a beacon
üAnalytics
➡ Access and view produced location data sets
2017-11-30 5Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th
6. üApplications
➡ Increasing awareness of
indoor navigation
➡ Working in wide range of
venues
➡ From simple proximity
notifications to interactive
navigation and asset tracking
in complex buildings
üExample customers
➡ High Point Market
• Yearly trade show with 75k
visitors in 6 buildings with 11
floors and over 1000 beacons
➡ Mumok
• Interactive tour guide app for
Vienna museum of modern art
➡ San Francisco Airport
• Navigation for blind users
Who our customers are?
2017-11-30 6Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th
7. Localization
How the blue dot is calculated.
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th
8. Radio fingerprinting localization.
üRadio reference map
➡ Point measurements of Signal strength
➡ Average several scans
➡ ~1 point / m2
➡ Must be up to date!
üBasic approach
➡ Find most similar points to observed
location
➡ Interpolate position from selected points
üFLIP
➡ FLexible Indoor Positioner
➡ Hierarchical approach
• Floor / Coarse / Fine
➡ Optimized reference data
➡ Handles differences in device
characteristics
2017-11-30 8
Beacon
Ref. point
Path Announcement
Est. position
Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th
9. Indoor Localization.
üHuman level accuracy
➡ ~2 m uncertainty
• AiLE - Advanced Indoo.rs Localization Engine
➡ ~2 s latency
• On device calculation
üLocalization
➡ Absolute position
• FLIP - Radio fingerprinting localization
• RSSI calibration independent
➡ Relative motion
• PDR - Pedestrian Dead Reckoning
• Orientation independent
• Step events with heading and length
➡ Trajectory
• AKF - Adaptive Kalman Filter
• Fuses absolute positions with relative motions
• Resistant to location jumps and heading drift
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th 9
FLIP PDR
AKF
Radio
Motion
sensors
Trajectory
StepsPositions
10. Radio maps
Crowd sourcing radio data.
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th
11. üChallenges
➡ Manual approach labor intensive
• Manual approach requires ~1 min / point, for ~1 point / m2!
• Large building needs tens of thousands of measurement points
➡ Noisy environment
➡ Radio maps require regular maintenance
• Beacons go missing, run out of battery, are moved
• Radio environment perturbed by new walls and objects
üSLAM: Three step approach – path to Big Data
➡ SLAM Engine
• Remove manual measurements
➡ SLAM Crowd Engine
• Crowd source data for updates
➡ Crowd Learning
• Remove dedicated measurements
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th 11
Radio mapping.
12. Initial map.
üSLAM Engine
➡ Use dedicated path recordings
• Occasional ground truths + PDR
➡ SLAM
• Graphic Model SLAM
• Global fit of positions and PDR
with radio constraints
➡ Gaussian interpolation
• Fuse measurements to fixed grid
üImplications
➡ 10-20x speedup
• Compared to manual approach
➡ Requires cloud calculations
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th 12
13. Map update.
üSLAM Crowd Engine
➡ Start with initial map
➡ Use crowd sourced paths
• Same data as for analytics
➡ SLAM
• Fuse estimated radio locations with PDR steps
üImprovement
➡ Heals and expands maps
➡ Removes redoing the initial map stage for
maintenance
üChallenges
➡ Requires a lot of data to overcome noise
➡ Difficult to make fully autonomous
➡ Requires frequent loops / path crossings
• Good for robots, less so for people...
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th 13
14. Map learning.
üCrowd learning
üGrow map from seeds
➡ Initially only basic navigation
➡ Seed sources
• GNNS, ray-tracing, proximity, partial map
üCrowd only based SLAM
➡ Join paths to close loops
üGrow and maintain map
➡ Reinforcement learning
➡ Fully automated
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th 14
15. SLAM evolution.
2017-11-30 15Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th
üWalk around
üUpdate maps
üTriggered
üPredefine path
üWalk path
üRepeat 10x
üWalk around
üCreate maps
üAutomatic
17. üData from analytics/crowd sourcing
➡ Detailed knowledge from navigating users available
➡ See crowding, obstacles, preferred routes, congestion etc.
Learn more from crowd data!
2017-11-30 17Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th
18. Building Utilization Intelligence.
üBUI goal
➡ Provide a complete picture of
how a building is used
➡ Combine building information
with localization data
➡ Provide tools to generate
insights from data set
üChallenges
➡ Dramatically increased data
volume
• Users are not always navigating
➡ New data sources
➡ Represent user mobility
patterns
➡ Go from static analytics to
knowledge discovery tool
üApproach
➡ Acquisition
• More sources
• ESRI ArcGIS
• IoT – motion detectors,
environment monitors, asset
tracking
• Usefulness assessment
➡ Integration
• Calibration
• Reliability estimation
➡ Data mining
• Aggregation into mobility spaces
➡ Interpretation
• Visual analytics
• API access
2017-11-30 18Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th
19. Building Utilization Intelligence.
2017-11-30 Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th 19
RESULTS AND BENEFITS
EXTERNAL INPUT
Data
Integration
Reliability
Estimation
IntegrationAcquisition Data Mining Interpretation
Knowledge
Discovery &
Monitoring
Prediction
Mobility Analysis
Mobile data
Infrastructure
Building
information
AnalystsMarket
Building utilization
intelligence
Accurate feature mapping
and localization
21. üindoo.rs enables indoor location
➡ SDK for third party apps
➡ Radio fingerprinting, pedestrian dead reckoning
üRadio maps are annoying
➡ Use crowd sourced data
➡ Big processing (rather than big data)
➡ Use SLAM to overcome noise
➡ Side effect: Cleaner analytics
üCrowd mobility
➡ Analytics beyond heat map / historic data
➡ Sufficient coverage requires heterogenous data
➡ Big Data
üPrediction is the future!
2017-11-30
Thomas Burgess <thomas@indoo.rs> 7th GeoIT Wherecamp Conference 2017, Berlin, Nov 30th
21
Summary and conclusions
22. indoor positioning and navigation for mobile apps
http://www.indoo.rs
Contact
7th GeoIT Wherecamp Conference 2017
Thomas Burgess
<thomas@indoo.rs>
tel:+43 720 11 5980
Thanks for
listening!