soqr: sonical quantifying the content level inside containers

23
SoQr: Sonical Quantifying the Content Level inside Containers Mingming Fan, Khai N. Truong UbiComp ‘15 1 Presenter: Ta Duc Tung @AKG - The University of Tokyo

Upload: tung-ta-duc

Post on 17-Feb-2017

170 views

Category:

Science


1 download

TRANSCRIPT

Page 1: SoQr: Sonical Quantifying the Content Level inside Containers

SoQr: Sonical Quantifying the Content Level inside

ContainersMingming Fan, Khai N. Truong

UbiComp ‘15

1

Presenter: Ta Duc Tung @AKG - The University of Tokyo

Page 2: SoQr: Sonical Quantifying the Content Level inside Containers

Outline• Introduction

• Related Works

• Theory of Operation

• System Design + Model Training

• Evaluation

• Practical Issues2

Page 3: SoQr: Sonical Quantifying the Content Level inside Containers

Introduction

Automate?3

Page 4: SoQr: Sonical Quantifying the Content Level inside Containers

Introduction

How much … is left? Milk, Juice

Washing Detergent

Eggs, Rice

4

Page 5: SoQr: Sonical Quantifying the Content Level inside Containers

Related Works

• Capacitive sensing

• Load sensing

• Camera sensing

• Electromagnetic Wave based sensing

• Acoustic sensing (time-of-flight)

5

Page 6: SoQr: Sonical Quantifying the Content Level inside Containers

Theory of OperationImpulse Response Analysis

6

Page 7: SoQr: Sonical Quantifying the Content Level inside Containers

System Design

7

Page 8: SoQr: Sonical Quantifying the Content Level inside Containers

System Design

8

Probing sound: 20Hz~20KHz sine wave

Record: 44.1KHz, 16 bits

Page 9: SoQr: Sonical Quantifying the Content Level inside Containers

System Design

9

Page 10: SoQr: Sonical Quantifying the Content Level inside Containers

Model Training

• Data collection• 0.01s sweep sound

• 1.01s response recording

• 100 samples for each level

10

Page 11: SoQr: Sonical Quantifying the Content Level inside Containers

Model Training

• Feature• Mel-Frequency Cepstral Coefficients - MFCCs

• Usually used in Speech Recognition

11

Page 12: SoQr: Sonical Quantifying the Content Level inside Containers

Model Training

• Model Learning - libSVM• Classification - Supported Vector Machines

• Regression - Supported Vector Regression

12

Page 13: SoQr: Sonical Quantifying the Content Level inside Containers

Evaluation

13

• Package:

• Material

• Is deformable?

• Shapes - Closable?

• Content:

• Type: liquid, solid, gel

• Density

Page 14: SoQr: Sonical Quantifying the Content Level inside Containers

Evaluation

• Most controlled environment - content at bottom

• Classification: overall F-Measure = 0.969

• Regression: mean absolute error = 0.0514

Page 15: SoQr: Sonical Quantifying the Content Level inside Containers

Evaluation• Most controlled environment - Eggs Package

15

Eggs Package

Page 16: SoQr: Sonical Quantifying the Content Level inside Containers

Evaluation

• Usage scenarios - Storing Places

• Train in refrigerator - Test on kitchen top

• Train on kitchen top - Test in refrigerator

• Overall Precision, Recall, F-Measure > 0.9

16

Page 17: SoQr: Sonical Quantifying the Content Level inside Containers

Evaluation• Usage scenarios - Container’s Orientation

• Separated training dataset

17

Stand-Up vs

Laid Down

Page 18: SoQr: Sonical Quantifying the Content Level inside Containers

Evaluation• Usage scenarios - Container’s Orientation

• Combined training dataset

18

Stand-Up vs

Laid Down

Page 19: SoQr: Sonical Quantifying the Content Level inside Containers

Evaluation• Usage scenarios - Deformable? Closable

• Deformable products

19

Potato Bag Opened - Clipped - Tied

Page 20: SoQr: Sonical Quantifying the Content Level inside Containers

Evaluation• Usage scenarios - Deformable? Closable

• Solid products - Closable

20

Page 21: SoQr: Sonical Quantifying the Content Level inside Containers

Evaluation

• Most controlled environment: GOOD

• Different storing places: GOOD

• Solid closable package: GOOD

• Different orientation: BAD

• Deformable package: BAD

21

Page 22: SoQr: Sonical Quantifying the Content Level inside Containers

Practical Issues

• Build prediction model for each product

• Sensor installing place

• Unconstrained context

22

Page 23: SoQr: Sonical Quantifying the Content Level inside Containers

Conclusion

• Small size sensor for content level sensing

• F-Measure > 0.96

• Sensing in different contexts

• Storing place, Orientation

• Rigid container

23