Towards Automatic Spatial Verification of Sensor Placement

17
Towards Automatic Spatial Verification of Sensor Placement Dezhi Hong Jorge Ortiz, Kamin Whitehouse, David Culler

description

Towards Automatic Spatial Verification of Sensor Placement. Dezhi Hong Jorge Ortiz, Kamin Whitehouse, David Culler. Why do we care?. Huge amount of sensor s , meters… Building setup changes Metadata management & maintenance Automated verification process . Before set off. - PowerPoint PPT Presentation

Transcript of Towards Automatic Spatial Verification of Sensor Placement

Page 1: Towards Automatic Spatial Verification of Sensor Placement

Towards Automatic Spatial Verification of Sensor Placement

Dezhi HongJorge Ortiz, Kamin Whitehouse, David Culler

Page 2: Towards Automatic Spatial Verification of Sensor Placement

Why do we care?

• Huge amount of sensors, meters…• Building setup changes• Metadata management & maintenance

Automated verification process

Page 3: Towards Automatic Spatial Verification of Sensor Placement

Before set off

• Statistical boundary?• Discoverability?• Convergence/Generalizability?

Page 4: Towards Automatic Spatial Verification of Sensor Placement

Methodology

• Empirical Mode Decomposition (EMD)• Intrinsic Mode Function (IMF) re-aggregation• Correlation analysis• Thresholding

Page 5: Towards Automatic Spatial Verification of Sensor Placement
Page 6: Towards Automatic Spatial Verification of Sensor Placement
Page 7: Towards Automatic Spatial Verification of Sensor Placement

IMF:(1) Same # of extrema and zero-crossings(2) Extrema symmetric to zero

Page 8: Towards Automatic Spatial Verification of Sensor Placement

Methodology• An example of EMD on a sensor trace

Page 9: Towards Automatic Spatial Verification of Sensor Placement

Methodology• IMF re-aggregation

2 temp. in diff. rms 2 sensors in a rm

Page 10: Towards Automatic Spatial Verification of Sensor Placement

Setup

• 5 rooms, 3 sensors/room• Sensor type: temperature, humidity, CO2

• Over a one-month period

Page 11: Towards Automatic Spatial Verification of Sensor Placement

Results

• Distribution generation

Page 12: Towards Automatic Spatial Verification of Sensor Placement

Results

• Receiver Operating Characteristic

• We choose the 0.2 FPR point as the boundary threshold for each room.

• TPR: 52%~93%, FPR: 5%~59%

On the mid IMF band On the raw traces

Page 13: Towards Automatic Spatial Verification of Sensor Placement

Results

• Convergence

• The threshold values converge to a similar value – 0.07

• Indicating generalizability

Page 14: Towards Automatic Spatial Verification of Sensor Placement

Results

• Clustering results (thresholding based)

14/15 correct = 93.3%

Page 15: Towards Automatic Spatial Verification of Sensor Placement

Results

• Clustering results (MDS + k-means)

On corrcoef from EMD-based

12/15 correct = 80%

On corrcoef from raw traces

8/15 correct = 53.3%

Page 16: Towards Automatic Spatial Verification of Sensor Placement

Conclusion

• A statistical boundary• Discoverable• Empirically generalizable

Page 17: Towards Automatic Spatial Verification of Sensor Placement

Qs?

Thank You