1. HydroSense: Infrastructure-Mediated Single-Point Sensing of Whole-Home Water Activity UbiComp 2009 Jon Froehlieh, Eric Larson, Tim Campbell, Conor Haggerty, James Fogarty, Shwetak N. Patel CS542 Internet system technology Presenter : Kim Young-sun

2. Introduction Important things in ubiquitous computing research and practice are… Effective methods for Sensing and Modeling human activity And requirements are… Practical / Low-cost / Unobtrusive

3. Sensing human activity Three approaches of sensing human activity - Mobile and wearable sensing - Environmental sensing - Infrastructure-mediated sensing

4. Sensing human activity Three approaches of sensing human activity - Mobile and wearable sensing - Environmental sensing - Infrastructure-mediated sensing

5. Sensing human activity Three approaches of sensing human activity - Mobile and wearable sensing - Environmental sensing - Infrastructure-mediated sensing

6. Sensing human activity Three approaches of sensing human activity - Mobile and wearable sensing - Environmental sensing - Infrastructure-mediated sensing water, electrical, and HVAC infrastructures

7. Infrastructure-mediated Sensing Especially, sensing of water activity E.g. Sensing from the Basement: A Feasibility Study of Unobtrusive and Low-Cost Home Activity Recognition

8. HydroSense A low-cost, single-point solution for activity sensing mediated by a home’s existing water infrastructure Advantages… - Easily installed - analysis of pressure - identify individual water fixtures - estimate the amount of water being used - diverse evaluation

9. Background In-home plumbing system

10. Background Pressure wave by valve open/close events Water Hammer

11. Background Estimating Flow Poiseuille’s law Fluid resistance formulation resulting in : Pressure Drop Radius of the pipe Length of the pipe Viscosity of the fluid

12. Prototype Sensor Design Prototype sensor implementation

13. In-home data collection Collected labeled data in – ten homes in four cities of varying, style, age, and diversity of plumbing system Measurement process –Measured the baseline static water pressure –Installed the appropriated HydroSense unit –Collect data * 706 fixture trials and 155 flow rate trials across 84 fixtures

14. In-home data collection A summary of the homes

15. Fixture event identification Three-step approach (1) Segment each individual valve event (2) Classify each valve event as either a valve open or valve close (3) Classify the valve event according to the individual fixture that generated it

16. Valve Event Segmentation Smooth raw signal using a low-pass linear phase finite impulse response filter Analyze in a sliding windows of 1000 samples (one second of sensed pressure) Check conditions of beginning and end of a valve event

17. Classifying Valve Events After segmenting each valve event, classify it as either a valve open or a valve close event Hierarchical classifier

18. Fixture Classification Associate valve open and valve close events with individual fixtures using a template-based hierarchical classifier Four complementary distance metrics –Matched filter –Matched derivative filter –Real Cepstrum –Mean squared error

19. Fixture Classification Similarity thresholds are learned from training data If no template passes all four filters, –The unknown event is not classified If templates corresponding to multiple fixtures pass all filters, –Choose among them using a nearest-neighbor classifier defined by the best performing distance metric, the matched derivative filter

20. Fixture Classification Evaluation Cross-validation experiment in different view

21. Analysis of flow estimation Individually Calibrated Valves

22. Analysis of flow estimation Generalizing to Uncalibrated Valves

23. Discussion Limitation and future work