11th International Conference on Mobile and Ubiquitous Systems: Computing, Network and Services – Dec 2nd-5th London F. Veronese, S. Comai, M. Matteucci, F. Salice Method, Design and Implementation of a Multiuser Indoor Localization System with Concurrent Fault Detection Fabio Veronese – fabio.veronese@polimi.it Biomedical Engineer, IT PhD Fellow ATG – Assistive Technology Group Dept. of Electronics Information and Bioengineering
AAL – Ambient Assisted Living Introduction Indoor Humans Localization LBS – Location Based Services ↓ Advertising Notification – Services… any? AAL – Ambient Assisted Living
Methods, Design and Implementation Outline Motivation Methods, Design and Implementation Experimental Results Simulations Conclusions Future Work
What are users asked to do? Usually IT approaches refer to: INVASIVENESS WEARABILITY MOBILITY BUT… … is the USER where we locate the device?
Dependability as a Target Indoor Humans Localization systems rarely provide a Dependable Service AAL Case Study implies: - User’s Needs Centered System Design - - Instrumented Environment -
Errors, Faults and Failures The result of localization can be not reliable WHY? → FAULTS! Natural Hardware - Human Made HOW can we DETECT them? Designing a Model of the systems Gathering extra data from another system
IHL systems provide an area where the person is with a certain error Modeling IHL Systems IHL systems provide an area where the person is with a certain error Let us consider the area centered in the estimation with a range based on the localization performances (3m ~ 84h percentile)
Gathering Information In AAL settings humans interact continuously, repeatedly and inevitably with Home Automation (HA) systems. →
Anonymous Interaction Detection Let us build a MODEL of interaction with DWS* What about More informative sensors? rs ( x,y ) Inactive No Information Toggled Person Interaction * Door Window Sensor
PIR* sensors have a more complicated behavior Presence Detection PIR* sensors have a more complicated behavior Inactive Active Inactive ( x,y ) rs No Movement Inactive Active Delay Active Inactive Persistence * Passive InfreRed
Presence Detection Model Let us define the behavior by cases: Entering and exiting a Sensor Area: Stopping in a Sensor Area Ideal Real Ideal Real
Concurrent Fault Detection Both systems have outputs based on the user position Indoor Humans Localization Human’s Real Position Fault Detection Apparatus Home Automation Sensor-detected Error Localization-detected Error
Fault Detection Policies The sets of possible locations from IHL and HA must allow a valid position All the HA activity must be motivated by a valid IHL position ? ?
Going MultiUser The handle multiuser we have to keep memory of the last IHL and HA valid position for each localized user ↓
Employed system were implemented based on AAL scenario settings: Who’s who? Employed system were implemented based on AAL scenario settings: Indoor Humans Localization Fault Detection Apparatus Home Automation
CAVEAT - Fault Observability “If a tree falls in the forest and no one is around to hear it… did it really fall?” If the user does not create a fault condition to produce an error which is visible then is the fault really happening?
Person Experimental Path Real World Tests TESTS No-fault Condition Natural Fault (Blinded Sensor) Human Fault (Forgotten Device) Sensor Positioning Person Experimental Path
Real World Results NO FAULT FORGOTTEN DEVICE SENSOR BLINDED
Making Real World Grow Limiting factors were making this not affordable: Few Devices Available Impossibility to test with several users Purchasing and Acquisition Times ↓ Large Environment Simulations: Unlimited Ideal Sensors Large Datasets Fast Generation
Simulated Tests TESTS No-fault Condition Natural Fault (Blinded Sensor) Human Fault (Forgotten Device)
Both Sensitivity and Specificity over 90% Simulations Results FORGOTTEN DEVICE NO FAULT BLINDED SENSOR Both Sensitivity and Specificity over 90%
Wrappin’ up The proposed method is based on the definition of a model representing a IHL and an HA systems, defining joint consistency conditions. The validity of the approach is proved applying it to a case study. The chosen case study subsystems are: LAURA localization system and a Z-wave based HA. The obtained experimental results showed the validity of our approach, correctly reporting errors in fault-free and fault injected conditions. Furthermore, we generated multiuser data, creating them based on the knowledge of the environment and the systems.
Conclusions Results of multiuser simulations show the system correctly detecting faults also in case of several targets. Both specificity and sensitivity above 90% represent a satisfying performance. Nonetheless, if the application requires it, system model parameters can be tuned to benefit selectively sensitivity or specificity. Concluding, our approach, even under some limitations in terms of fault observability, enables the dependable localization of a set of persons inside an instrumented house, detecting both natural and human-made faults.
Future Work GROWTH Real World installation Pattern Identification Fault Identification and Localization
Thanks Fabio Veronese – fabio.veronese@polimi.it Biomedical Engineer, IT PhD Fellow ATG – Assistive Technology Group Dept. of Electronics Information and Bioengineering