AUTOMATED BUILDING ENERGY CONSUMPTION ESTIMATION FROM AERIAL IMAGERY - - PowerPoint PPT Presentation

▶
automated building energy consumption estimation from
SMART_READER_LITE
LIVE PREVIEW

AUTOMATED BUILDING ENERGY CONSUMPTION ESTIMATION FROM AERIAL IMAGERY - - PowerPoint PPT Presentation

Sep 29, 2023 1.31k likes •1.88k views

AUTOMATED BUILDING ENERGY CONSUMPTION ESTIMATION FROM AERIAL IMAGERY A BASS CONNECTIONS IN ENERGY PROJECT TEAM Student Researchers Mitchell Kim Sebastian Lin Sophia Park Eric Peshkin Pratt 18 Trinity 18 Pratt 17 Trinity 18 T

slide-1
SLIDE 1

AUTOMATED BUILDING ENERGY CONSUMPTION ESTIMATION FROM AERIAL IMAGERY

A BASS CONNECTIONS IN ENERGY PROJECT TEAM

slide-2
SLIDE 2

2

Mitchell Kim

Pratt ‘18

Sophia Park

Pratt ‘17

Sebastian Lin

Trinity ‘18

Eric Peshkin

Trinity ‘18

Nikhil Vanderklaauw

Pratt ‘18

Yue Xi

Trinity ‘19

Hoël Wiesner

Nicholas ‘17

Samit Sura

Economics ‘17

  • Dr. Timothy Johnson

Nicholas School

  • Dr. Kyle Bradbury

Energy Initiative

  • Dr. Leslie Collins

Pratt School

Student Researchers Faculty Advisors

T eam Members 2016-17

slide-3
SLIDE 3

AUTOMATED BUILDING ENERGY CONSUMPTION ESTIMATION FROM AERIAL IMAGERY

A BASS CONNECTIONS IN ENERGY PROJECT TEAM

slide-4
SLIDE 4

What can an image tell us about our energy consumption?

slide-5
SLIDE 5
slide-6
SLIDE 6
slide-7
SLIDE 7
slide-8
SLIDE 8

8

Governments and policy makers

slide-9
SLIDE 9

9

Governments and policy makers Businesses and NGOs

slide-10
SLIDE 10

10

Governments and policy makers Businesses and NGOs Researchers

slide-11
SLIDE 11

11

Our Process

slide-12
SLIDE 12

12

From a high resolution aerial image…

slide-13
SLIDE 13

13

Detect building

  • utlines and

calculate their area

slide-14
SLIDE 14

14

slide-15
SLIDE 15

15

Use area of detected buildings for energy use estimation

slide-16
SLIDE 16

16

slide-17
SLIDE 17

17

slide-18
SLIDE 18

18

Approach 1: Random Forests Approach 2: Convolutional Neural Network

slide-19
SLIDE 19

19

Approach 1: Random Forests Approach 2: Convolutional Neural Network Evaluate and Compare Results

slide-20
SLIDE 20

20

Approach 1: Random Forests Approach 2: Convolutional Neural Network Evaluate and Compare Results

Select Building Detection Approach

slide-21
SLIDE 21

21

How Can We “Teach” a Computer?

Learning Algorithm (Random Forest) Detected Buildings Feature Extraction Classification

Approach 1: Classical Machine Learning

slide-22
SLIDE 22

22

Approach 1: Classical Machine Learning

Features:

slide-23
SLIDE 23

23

Approach 1: Classical Machine Learning

Features: § Color Data (HSV)

slide-24
SLIDE 24

24

Approach 1: Classical Machine Learning

Features: § Color Data (HSV) § Edges (Gradient)

slide-25
SLIDE 25

25

Approach 1: Classical Machine Learning

Features: § Color Data (HSV) § Edges (Gradient) § Variation in Pixels (STDev)

slide-26
SLIDE 26

26

Approach 1: Classical Machine Learning

Features: § Color Data (HSV) § Edges (Gradient) § Variation in Pixels (STDev) § T exture (Entropy)

slide-27
SLIDE 27

27

Approach 1: Classical Machine Learning

Features: § Color Data (HSV) § Edges (Gradient) § Variation in Pixels (STDev) § T exture (Entropy) § Vegetation Detection (NDVI)

slide-28
SLIDE 28

YES NO Color? Shape? Texture? Rectangular Non- Rectangular Coarse Smooth Gray Green YES NO NO NO

Approach 1: Classical Machine Learning

Decision Tree: Question: Is the pixel part of a building? Answer:

28

slide-29
SLIDE 29

29

Approach 1: Classical Machine Learning

Random Forest YES YES YES NO NO YES Vote = Input Pixel

slide-30
SLIDE 30

30

Approach II: Convolutional Neural Network

slide-31
SLIDE 31

31

Approach II: Convolutional Neural Network

slide-32
SLIDE 32

32

Approach II: Convolutional Neural Network

slide-33
SLIDE 33

33

Neural Network vs. Random Forest Classifier

Features?

Time?

Approach II: Convolutional Neural Network

slide-34
SLIDE 34

Pool Building Car Tennis Court Building Car Pool Court

34

Our Neural Network: Overview

Adapted From: https://www.mathworks.com/help/nnet/convolutional-neural-networks.html

Approach II: Convolutional Neural Network

building

slide-35
SLIDE 35

35

Comparing Approaches:

Ground truth building outlines, i.e., the ideal classification output Building outlines detected by random forest classification Building outlines detected by convolutional neural network

slide-36
SLIDE 36

36

Comparing Approaches:

Ground truth building outlines, i.e., the ideal classification output Building outlines detected by random forest classification Building outlines detected by convolutional neural network Misclassified building pixel "islands"

slide-37
SLIDE 37

37

Comparing Approaches:

Ground truth building outlines, i.e., the ideal classification output Building outlines detected by random forest classification Building outlines detected by convolutional neural network Irregular edges & merged buildings

slide-38
SLIDE 38

38

Comparing Approaches:

slide-39
SLIDE 39

39

Our Process

slide-40
SLIDE 40

40

How good is the model?

slide-41
SLIDE 41

41

How good is the model?

slide-42
SLIDE 42

Actual Buildings and Energy Consumption

Number of Buildings Average Energy Use (kWh/yr) T

  • tal Energy

Estimation Error (%)

388 10,237

slide-43
SLIDE 43

Actual Buildings and Energy Consumption Actual Buildings and Estimated Energy Consumption

Number of Buildings Average Energy Use (kWh/yr) T

  • tal Energy

Estimation Error (%)

388 10,237

  • 388

11,977 17%

slide-44
SLIDE 44

Actual Buildings and Energy Consumption Actual Buildings and Estimated Energy Consumption Detected Buildings and Estimated Energy Consumption

Number of Buildings Average Energy Use (kWh/yr) T

  • tal Energy

Estimation Error (%)

388 10,237

  • 388

11,977 17% 299 12,405

  • 7%
slide-45
SLIDE 45

45

Conclusion

From a high resolution aerial image…

slide-46
SLIDE 46

46

Conclusion

From a high resolution aerial image… Detect building

  • utlines and

calculate their area

slide-47
SLIDE 47

47

Conclusion

From a high resolution aerial image… Detect building

  • utlines and

calculate their area Use area of detected buildings for energy use estimation

slide-48
SLIDE 48

48

Conclusion

Scale up to gather this data for whole cities, with thousands of buildings, anywhere in the world!

slide-49
SLIDE 49

Solving Murders!

slide-50
SLIDE 50
slide-51
SLIDE 51
slide-52
SLIDE 52
slide-53
SLIDE 53
slide-54
SLIDE 54
slide-55
SLIDE 55

And even winning awards for presenting!

slide-56
SLIDE 56

56

THANK YOU