SlideShare a Scribd company logo
Jui-Hsin(Larry) Lai, Chung-Ching Lin,
 

Chun-Fu(Richard) Chen, and Ching-Yung Li
n

IBM T.J. Watson Research Center
Multi-Modality Mobile Image
Recognition Based on
Thermal and Visual Cameras
Demo Videos


www.larry-lai.com/mobilethermal.html
IBM :: IBM Confidential :: ©2015 IBM Corporation 2
IBM :: IBM Confidential :: ©2015 IBM Corporation
Expert Tech APP with SmallBlue
3
• SmallBlue


• find your colleagues with the
knowledge that your are looking for
• Expert Tech APP


• give field technicians the right tools
for the job of customer service
IBM :: IBM Confidential :: ©2015 IBM Corporation
Demo: Expert Tech APP with SmallBlue
4
IBM :: IBM Confidential :: ©2015 IBM Corporation
More Functions in Expert Tech APP
5
• Expert Tech APP


• give field technicians the right tools
for the job of customer service
A trouble to workers— when facing a device, technicians have no
idea/experience to fix it. How can Expert Tech do the help?
IBM :: IBM Confidential :: ©2015 IBM Corporation
The Help from Image Recognition
6
• Where the computation take place


• running image recognition on the cloud server while it
connects to Internet


• running image recognition on the mobile device while there
is no Internet connection
• Steps


• (1) take a picture of an asset,


• (2) run image recognition algorithm,


• (3) and show the instructions to fix it step-by-step
IBM :: IBM Confidential :: ©2015 IBM Corporation
Image Recognition Running on Mobile Devices
7
• Challenges


• low computation capability on mobile devices


• require immediate response for real-time applications
• Computation Acceleration


• Feature detection: Harris Corner Detector is implemented
because it can be easily modified into pixel-wise operation


=> by leveraging CIKernal, the iOS image processing
framework, to use GPU’s for calculation


• Feature description: modify SIFT algorithm to reduce feature
vectors


=> by leveraging multi-thread for acceleration


• Feature matching: image features are stored with CoreDate,
the iOS SQL framework, for efficient data access
IBM :: IBM Confidential :: ©2015 IBM Corporation
Not Only Recognition, But Appearance Inspection
8
Cancel
Check the power signal
The LED should be green light as
a normal condition.
View instructions
Fixing Point 2
• Geometry Mapping & Local Feature Extraction


• Register image features for assets under normal conditions,
like the light color, component shape


• Compare features of captured images to the database
IBM :: IBM Confidential :: ©2015 IBM Corporation
Expert Tech APP with Image Recognition
9
IBM :: IBM Confidential :: ©2015 IBM Corporation
Do More Beyond Human Eye
• Asset Temperature Readings


• Take thermal and visible images of an asset, run image recognition to
identify the asset’s model, and automatically pull temperature readings
to check asset’s condition


• Show instructions to fix it and recommend the experts to help
10
• Identify a Heat Source


• Point the camera at an asset, analyze the real-time video stream, and
overlay the image with indicators showing the hear source


• Analyze the heat pattern change on temporal domain => computation
in real-time
IBM :: IBM Confidential :: ©2015 IBM Corporation
Do More Beyond Human Eye
11
• Image recognition with Visual and Thermal Cameras


• leverage mobile computation for real-time applications
Visible Image
from smartphone
Thermal and visible
images from external
camera
Smartphone with
external camera
IBM :: IBM Confidential :: ©2015 IBM Corporation
Image Overlap — Thermal and Visual Images
12
(b)Thermal image (c)Visible image
(d)Overlapping thermal
and visible images
without displacement
(a)External camera
• Problem: the mismatch in overlapping images
IBM :: IBM Confidential :: ©2015 IBM Corporation
Image Overlap — Thermal and Visual Images
13
(e)Overlapping with
displacement d1 to focus
on the closer mug
(f)Overlapping with
displacement d2 to focus
on the farther mug
• Different object in different depth has its own displacement d
IBM :: IBM Confidential :: ©2015 IBM Corporation
Thermal-Visible Image Alignment
14
• The image texture is different


• Conventional image registration using key features is not working


• An observation: temperature distribution is correlated to object shape


• => Minimize the difference between edge maps
Et(x,y) is the edge map of thermal image
 

E ̃d(x,y) is the edge map of visible image
with a displacement d
IBM :: IBM Confidential :: ©2015 IBM Corporation
Demo — Thermal-Visual Image Alignment
15
IBM :: IBM Confidential :: ©2015 IBM Corporation
Asset Inspection with Thermal and Visual Images
16
IBM :: IBM Confidential :: ©2015 IBM Corporation
Heat Source and Pattern Detection
17
frame m
frame m+1
frame n frame t
frame n+1 frame t+1
(a)Lighting source (b)Spot heating source (c)Strip heating source
• An application scenario


• field workers would like to inspect heating system and understand the
heat distribution in every corner of a house


• predefine some common patterns with distinguishable traits
IBM :: IBM Confidential :: ©2015 IBM Corporation
Detection Criteria for Heat Patterns
18
• Light Source


• A region with higher temperature than its surroundings is detected,
whose luminance distribution on visible video frames is high as well
• Spot Heating Source


• A region with higher temperature than its surroundings is detected, whose
distribution like a circle and stable over a period of time, and the luminance
distribution on visible video frames is not higher than its surroundings.
• Strip Heating Source


• A region with higher temperature than its surroundings is detected, whose
distribution like a strip and stable over a period of time, and the luminance
distribution on visible video frames is not higher than its surroundings
IBM :: IBM Confidential :: ©2015 IBM Corporation
Screenshot of Heat Source Detection and Pattern Analysis
19
Cancel Heat Camera List
1
3
2
Cancel
Spot Heat
Detect a spot heat with 135 F in the
center. Heat flow slightly goes up.
Checking Point 2
IBM :: IBM Confidential :: ©2015 IBM Corporation
Conclusion
20
• Applications with visible and thermal image processing


• image recognition for helping field workers identify asset model,
inspect asset condition, and show instructions step-by-step


• thermal and visual image processing for heat pattern detection and
operation condition check
• Mobile analytics is a huge demand and can help users
to solve problems
• More and more mobile analytics on the way


• viewpoints from research to product


• viewpoints from demand to research
IBM :: IBM Confidential :: ©2015 IBM Corporation
Thanks for Your Attention
21

More Related Content

Similar to Multi-Modality Mobile Image Recognition Based on Thermal and Visual Cameras

G-net - Company Profile
G-net - Company ProfileG-net - Company Profile
G-net - Company Profile
Julladaj Bleriot
 
NET New Electronic Technology GmbH - New Potential of embedded vision systems...
NET New Electronic Technology GmbH - New Potential of embedded vision systems...NET New Electronic Technology GmbH - New Potential of embedded vision systems...
NET New Electronic Technology GmbH - New Potential of embedded vision systems...
Kurt Buttress
 
BigML Release: Image Processing
BigML Release: Image ProcessingBigML Release: Image Processing
BigML Release: Image Processing
BigML, Inc
 
Image Processing Basics
Image Processing BasicsImage Processing Basics
Image Processing Basics
A B Shinde
 
Banish PCB Thermal Problems with State-of-the-art Thermal Simulation
Banish PCB Thermal Problems with State-of-the-art Thermal SimulationBanish PCB Thermal Problems with State-of-the-art Thermal Simulation
Banish PCB Thermal Problems with State-of-the-art Thermal Simulation
Tom Gregory
 
Improving Quality & Profits with Vision
Improving Quality & Profits with VisionImproving Quality & Profits with Vision
Improving Quality & Profits with VisionTim Seymour
 
DMC NI Week 2014 High Speed Vision
DMC NI Week 2014 High Speed VisionDMC NI Week 2014 High Speed Vision
DMC NI Week 2014 High Speed Vision
DMC, Inc.
 
Mx Presentation En 2008
Mx Presentation En 2008Mx Presentation En 2008
Mx Presentation En 2008
Nour Al-Shakhouri
 
IEEE EED2021 AI use cases in Computer Vision
IEEE EED2021 AI use cases in Computer VisionIEEE EED2021 AI use cases in Computer Vision
IEEE EED2021 AI use cases in Computer Vision
SAMeh Zaghloul
 
1040 ibm worklight delivering agility to mobile cloud deployments
1040 ibm worklight  delivering agility to mobile cloud deployments1040 ibm worklight  delivering agility to mobile cloud deployments
1040 ibm worklight delivering agility to mobile cloud deploymentsTodd Kaplinger
 
Building a Custom Camera Application in Android
Building a Custom Camera Application in AndroidBuilding a Custom Camera Application in Android
Building a Custom Camera Application in Android
Huyen Dao
 
AnDevCon 2014: Building a Custom Camera Application
AnDevCon 2014: Building a Custom Camera ApplicationAnDevCon 2014: Building a Custom Camera Application
AnDevCon 2014: Building a Custom Camera Application
Huyen Tue Dao
 
Digital Image Processing - Reprsentation
Digital Image Processing - ReprsentationDigital Image Processing - Reprsentation
Digital Image Processing - Reprsentation
ssuser7ec6af
 
Repro with Ipro: Simplifying your Imaging Workflows
Repro with Ipro: Simplifying your Imaging WorkflowsRepro with Ipro: Simplifying your Imaging Workflows
Repro with Ipro: Simplifying your Imaging Workflows
Ipro Tech
 
“Selecting the Right Camera for Your Embedded Computer Vision Project,” a Pre...
“Selecting the Right Camera for Your Embedded Computer Vision Project,” a Pre...“Selecting the Right Camera for Your Embedded Computer Vision Project,” a Pre...
“Selecting the Right Camera for Your Embedded Computer Vision Project,” a Pre...
Edge AI and Vision Alliance
 
BTLE (Bluetooth Low Energy) and CoreBluetooth
BTLE (Bluetooth Low Energy) and CoreBluetooth BTLE (Bluetooth Low Energy) and CoreBluetooth
BTLE (Bluetooth Low Energy) and CoreBluetooth
Zach Dennis
 
"Selecting the Right Imager for Your Embedded Vision Application," a Presenta...
"Selecting the Right Imager for Your Embedded Vision Application," a Presenta..."Selecting the Right Imager for Your Embedded Vision Application," a Presenta...
"Selecting the Right Imager for Your Embedded Vision Application," a Presenta...
Edge AI and Vision Alliance
 
"The Coming Shift from Image Sensors to Image Sensing," a Presentation from LG
"The Coming Shift from Image Sensors to Image Sensing," a Presentation from LG"The Coming Shift from Image Sensors to Image Sensing," a Presentation from LG
"The Coming Shift from Image Sensors to Image Sensing," a Presentation from LG
Edge AI and Vision Alliance
 
Computer architecture for vision system
Computer architecture for vision systemComputer architecture for vision system
Computer architecture for vision system
AkashPatil334
 
"Deep Learning for Manufacturing Inspection Applications," a Presentation fro...
"Deep Learning for Manufacturing Inspection Applications," a Presentation fro..."Deep Learning for Manufacturing Inspection Applications," a Presentation fro...
"Deep Learning for Manufacturing Inspection Applications," a Presentation fro...
Edge AI and Vision Alliance
 

Similar to Multi-Modality Mobile Image Recognition Based on Thermal and Visual Cameras (20)

G-net - Company Profile
G-net - Company ProfileG-net - Company Profile
G-net - Company Profile
 
NET New Electronic Technology GmbH - New Potential of embedded vision systems...
NET New Electronic Technology GmbH - New Potential of embedded vision systems...NET New Electronic Technology GmbH - New Potential of embedded vision systems...
NET New Electronic Technology GmbH - New Potential of embedded vision systems...
 
BigML Release: Image Processing
BigML Release: Image ProcessingBigML Release: Image Processing
BigML Release: Image Processing
 
Image Processing Basics
Image Processing BasicsImage Processing Basics
Image Processing Basics
 
Banish PCB Thermal Problems with State-of-the-art Thermal Simulation
Banish PCB Thermal Problems with State-of-the-art Thermal SimulationBanish PCB Thermal Problems with State-of-the-art Thermal Simulation
Banish PCB Thermal Problems with State-of-the-art Thermal Simulation
 
Improving Quality & Profits with Vision
Improving Quality & Profits with VisionImproving Quality & Profits with Vision
Improving Quality & Profits with Vision
 
DMC NI Week 2014 High Speed Vision
DMC NI Week 2014 High Speed VisionDMC NI Week 2014 High Speed Vision
DMC NI Week 2014 High Speed Vision
 
Mx Presentation En 2008
Mx Presentation En 2008Mx Presentation En 2008
Mx Presentation En 2008
 
IEEE EED2021 AI use cases in Computer Vision
IEEE EED2021 AI use cases in Computer VisionIEEE EED2021 AI use cases in Computer Vision
IEEE EED2021 AI use cases in Computer Vision
 
1040 ibm worklight delivering agility to mobile cloud deployments
1040 ibm worklight  delivering agility to mobile cloud deployments1040 ibm worklight  delivering agility to mobile cloud deployments
1040 ibm worklight delivering agility to mobile cloud deployments
 
Building a Custom Camera Application in Android
Building a Custom Camera Application in AndroidBuilding a Custom Camera Application in Android
Building a Custom Camera Application in Android
 
AnDevCon 2014: Building a Custom Camera Application
AnDevCon 2014: Building a Custom Camera ApplicationAnDevCon 2014: Building a Custom Camera Application
AnDevCon 2014: Building a Custom Camera Application
 
Digital Image Processing - Reprsentation
Digital Image Processing - ReprsentationDigital Image Processing - Reprsentation
Digital Image Processing - Reprsentation
 
Repro with Ipro: Simplifying your Imaging Workflows
Repro with Ipro: Simplifying your Imaging WorkflowsRepro with Ipro: Simplifying your Imaging Workflows
Repro with Ipro: Simplifying your Imaging Workflows
 
“Selecting the Right Camera for Your Embedded Computer Vision Project,” a Pre...
“Selecting the Right Camera for Your Embedded Computer Vision Project,” a Pre...“Selecting the Right Camera for Your Embedded Computer Vision Project,” a Pre...
“Selecting the Right Camera for Your Embedded Computer Vision Project,” a Pre...
 
BTLE (Bluetooth Low Energy) and CoreBluetooth
BTLE (Bluetooth Low Energy) and CoreBluetooth BTLE (Bluetooth Low Energy) and CoreBluetooth
BTLE (Bluetooth Low Energy) and CoreBluetooth
 
"Selecting the Right Imager for Your Embedded Vision Application," a Presenta...
"Selecting the Right Imager for Your Embedded Vision Application," a Presenta..."Selecting the Right Imager for Your Embedded Vision Application," a Presenta...
"Selecting the Right Imager for Your Embedded Vision Application," a Presenta...
 
"The Coming Shift from Image Sensors to Image Sensing," a Presentation from LG
"The Coming Shift from Image Sensors to Image Sensing," a Presentation from LG"The Coming Shift from Image Sensors to Image Sensing," a Presentation from LG
"The Coming Shift from Image Sensors to Image Sensing," a Presentation from LG
 
Computer architecture for vision system
Computer architecture for vision systemComputer architecture for vision system
Computer architecture for vision system
 
"Deep Learning for Manufacturing Inspection Applications," a Presentation fro...
"Deep Learning for Manufacturing Inspection Applications," a Presentation fro..."Deep Learning for Manufacturing Inspection Applications," a Presentation fro...
"Deep Learning for Manufacturing Inspection Applications," a Presentation fro...
 

More from Jui-Hsin (Larry) Lai

20211118 AI+ Remote Sensing
20211118 AI+ Remote Sensing20211118 AI+ Remote Sensing
20211118 AI+ Remote Sensing
Jui-Hsin (Larry) Lai
 
Theme-Matters: Fashion Compatibility Learning via Theme Attention
Theme-Matters: Fashion Compatibility Learning via Theme AttentionTheme-Matters: Fashion Compatibility Learning via Theme Attention
Theme-Matters: Fashion Compatibility Learning via Theme Attention
Jui-Hsin (Larry) Lai
 
Tennis Real Play
Tennis Real PlayTennis Real Play
Tennis Real Play
Jui-Hsin (Larry) Lai
 
Object Tracking with Instance Matching and Online Learning
Object Tracking with Instance Matching and Online LearningObject Tracking with Instance Matching and Online Learning
Object Tracking with Instance Matching and Online Learning
Jui-Hsin (Larry) Lai
 
Theme-Matters: Fashion Compatibility Learning via Theme Attention
Theme-Matters: Fashion Compatibility Learning via Theme AttentionTheme-Matters: Fashion Compatibility Learning via Theme Attention
Theme-Matters: Fashion Compatibility Learning via Theme Attention
Jui-Hsin (Larry) Lai
 
AI+ Remote Sensing: Applying Deep Learning to Image Enhancement, Analytics, a...
AI+ Remote Sensing: Applying Deep Learning to Image Enhancement, Analytics, a...AI+ Remote Sensing: Applying Deep Learning to Image Enhancement, Analytics, a...
AI+ Remote Sensing: Applying Deep Learning to Image Enhancement, Analytics, a...
Jui-Hsin (Larry) Lai
 

More from Jui-Hsin (Larry) Lai (6)

20211118 AI+ Remote Sensing
20211118 AI+ Remote Sensing20211118 AI+ Remote Sensing
20211118 AI+ Remote Sensing
 
Theme-Matters: Fashion Compatibility Learning via Theme Attention
Theme-Matters: Fashion Compatibility Learning via Theme AttentionTheme-Matters: Fashion Compatibility Learning via Theme Attention
Theme-Matters: Fashion Compatibility Learning via Theme Attention
 
Tennis Real Play
Tennis Real PlayTennis Real Play
Tennis Real Play
 
Object Tracking with Instance Matching and Online Learning
Object Tracking with Instance Matching and Online LearningObject Tracking with Instance Matching and Online Learning
Object Tracking with Instance Matching and Online Learning
 
Theme-Matters: Fashion Compatibility Learning via Theme Attention
Theme-Matters: Fashion Compatibility Learning via Theme AttentionTheme-Matters: Fashion Compatibility Learning via Theme Attention
Theme-Matters: Fashion Compatibility Learning via Theme Attention
 
AI+ Remote Sensing: Applying Deep Learning to Image Enhancement, Analytics, a...
AI+ Remote Sensing: Applying Deep Learning to Image Enhancement, Analytics, a...AI+ Remote Sensing: Applying Deep Learning to Image Enhancement, Analytics, a...
AI+ Remote Sensing: Applying Deep Learning to Image Enhancement, Analytics, a...
 

Multi-Modality Mobile Image Recognition Based on Thermal and Visual Cameras

  • 1. Jui-Hsin(Larry) Lai, Chung-Ching Lin, Chun-Fu(Richard) Chen, and Ching-Yung Li n IBM T.J. Watson Research Center Multi-Modality Mobile Image Recognition Based on Thermal and Visual Cameras Demo Videos www.larry-lai.com/mobilethermal.html
  • 2. IBM :: IBM Confidential :: ©2015 IBM Corporation 2
  • 3. IBM :: IBM Confidential :: ©2015 IBM Corporation Expert Tech APP with SmallBlue 3 • SmallBlue • find your colleagues with the knowledge that your are looking for • Expert Tech APP • give field technicians the right tools for the job of customer service
  • 4. IBM :: IBM Confidential :: ©2015 IBM Corporation Demo: Expert Tech APP with SmallBlue 4
  • 5. IBM :: IBM Confidential :: ©2015 IBM Corporation More Functions in Expert Tech APP 5 • Expert Tech APP • give field technicians the right tools for the job of customer service A trouble to workers— when facing a device, technicians have no idea/experience to fix it. How can Expert Tech do the help?
  • 6. IBM :: IBM Confidential :: ©2015 IBM Corporation The Help from Image Recognition 6 • Where the computation take place • running image recognition on the cloud server while it connects to Internet • running image recognition on the mobile device while there is no Internet connection • Steps • (1) take a picture of an asset, • (2) run image recognition algorithm, • (3) and show the instructions to fix it step-by-step
  • 7. IBM :: IBM Confidential :: ©2015 IBM Corporation Image Recognition Running on Mobile Devices 7 • Challenges • low computation capability on mobile devices • require immediate response for real-time applications • Computation Acceleration • Feature detection: Harris Corner Detector is implemented because it can be easily modified into pixel-wise operation 
 => by leveraging CIKernal, the iOS image processing framework, to use GPU’s for calculation • Feature description: modify SIFT algorithm to reduce feature vectors 
 => by leveraging multi-thread for acceleration • Feature matching: image features are stored with CoreDate, the iOS SQL framework, for efficient data access
  • 8. IBM :: IBM Confidential :: ©2015 IBM Corporation Not Only Recognition, But Appearance Inspection 8 Cancel Check the power signal The LED should be green light as a normal condition. View instructions Fixing Point 2 • Geometry Mapping & Local Feature Extraction • Register image features for assets under normal conditions, like the light color, component shape • Compare features of captured images to the database
  • 9. IBM :: IBM Confidential :: ©2015 IBM Corporation Expert Tech APP with Image Recognition 9
  • 10. IBM :: IBM Confidential :: ©2015 IBM Corporation Do More Beyond Human Eye • Asset Temperature Readings • Take thermal and visible images of an asset, run image recognition to identify the asset’s model, and automatically pull temperature readings to check asset’s condition • Show instructions to fix it and recommend the experts to help 10 • Identify a Heat Source • Point the camera at an asset, analyze the real-time video stream, and overlay the image with indicators showing the hear source • Analyze the heat pattern change on temporal domain => computation in real-time
  • 11. IBM :: IBM Confidential :: ©2015 IBM Corporation Do More Beyond Human Eye 11 • Image recognition with Visual and Thermal Cameras • leverage mobile computation for real-time applications Visible Image from smartphone Thermal and visible images from external camera Smartphone with external camera
  • 12. IBM :: IBM Confidential :: ©2015 IBM Corporation Image Overlap — Thermal and Visual Images 12 (b)Thermal image (c)Visible image (d)Overlapping thermal and visible images without displacement (a)External camera • Problem: the mismatch in overlapping images
  • 13. IBM :: IBM Confidential :: ©2015 IBM Corporation Image Overlap — Thermal and Visual Images 13 (e)Overlapping with displacement d1 to focus on the closer mug (f)Overlapping with displacement d2 to focus on the farther mug • Different object in different depth has its own displacement d
  • 14. IBM :: IBM Confidential :: ©2015 IBM Corporation Thermal-Visible Image Alignment 14 • The image texture is different • Conventional image registration using key features is not working • An observation: temperature distribution is correlated to object shape • => Minimize the difference between edge maps Et(x,y) is the edge map of thermal image E ̃d(x,y) is the edge map of visible image with a displacement d
  • 15. IBM :: IBM Confidential :: ©2015 IBM Corporation Demo — Thermal-Visual Image Alignment 15
  • 16. IBM :: IBM Confidential :: ©2015 IBM Corporation Asset Inspection with Thermal and Visual Images 16
  • 17. IBM :: IBM Confidential :: ©2015 IBM Corporation Heat Source and Pattern Detection 17 frame m frame m+1 frame n frame t frame n+1 frame t+1 (a)Lighting source (b)Spot heating source (c)Strip heating source • An application scenario • field workers would like to inspect heating system and understand the heat distribution in every corner of a house • predefine some common patterns with distinguishable traits
  • 18. IBM :: IBM Confidential :: ©2015 IBM Corporation Detection Criteria for Heat Patterns 18 • Light Source • A region with higher temperature than its surroundings is detected, whose luminance distribution on visible video frames is high as well • Spot Heating Source • A region with higher temperature than its surroundings is detected, whose distribution like a circle and stable over a period of time, and the luminance distribution on visible video frames is not higher than its surroundings. • Strip Heating Source • A region with higher temperature than its surroundings is detected, whose distribution like a strip and stable over a period of time, and the luminance distribution on visible video frames is not higher than its surroundings
  • 19. IBM :: IBM Confidential :: ©2015 IBM Corporation Screenshot of Heat Source Detection and Pattern Analysis 19 Cancel Heat Camera List 1 3 2 Cancel Spot Heat Detect a spot heat with 135 F in the center. Heat flow slightly goes up. Checking Point 2
  • 20. IBM :: IBM Confidential :: ©2015 IBM Corporation Conclusion 20 • Applications with visible and thermal image processing • image recognition for helping field workers identify asset model, inspect asset condition, and show instructions step-by-step • thermal and visual image processing for heat pattern detection and operation condition check • Mobile analytics is a huge demand and can help users to solve problems • More and more mobile analytics on the way • viewpoints from research to product • viewpoints from demand to research
  • 21. IBM :: IBM Confidential :: ©2015 IBM Corporation Thanks for Your Attention 21