SlideShare a Scribd company logo
Smart Vision System:
Design of Application
Algorithm-Architecture
      Eric Delfosse
      IBBT-NES-IMEC
What are Smart Vision Systems?

Systems that embed intelligence through advanced
image processing to:

  Enhance visual user experience

  Improve natural interaction

  Facilitate decision making for complex events

  …
Enhance visual user experience
Multi-camera 3D image reconstruction for advanced
surveillance




3D GPS
                                           (IBBT GBO ISYSS)


                                           (IBBT GBO URBAN)
Increase natural interaction


Natural human-machine interface
through gesture recognition




3D (immersive) video              (IBBT GBO Hi-Masquerade)
conferencing
Facilitate decision making for complex events


Event detection for
surveillance applications


                        (IBBT GBO ISYSS)




Traffic sign recognition for
driver assistance

                       (IBBT GBO URBAN)
Different applications, different requirements

                   Throughput
                  High resolution
                  High framerate
Functionality
                                          Accuracy
                                    Reduce false negatives
                                      and false positives


                      …
          Size                         Low Power
                                       Solar energy
                                    Avoid active cooling
                                             ...

                         Cost
These applications require increasingly complex
algorithms

                                                    Exponential algorithmic complexity increase
                                                                                                                                          > 9000
                            3000


                            2500
   Complexity (Ops/pixel)




                            2000


                            1500


                            1000


                             500


                               0
                                                                       enhancement




                                                                                       extraction




                                                                                                    Depth extraction
                                   Edge detection




                                                        AVC encoding




                                                                                                                       Object detection




                                                                                                                                           understanding
                                                                                        Feature
                                                                          Image




                                                                                                                                              Image
                                                                                     Algorithm
Diversity of platform architectures with different
characteristics




                                   PC + GPU
                                 High performance
                                    High power
                             Limited portability (laptop)
                                         …




      Server rack
                                Embedded systems
    Very high performance
       Very high Power          Medium performance
         Non-portable               Low power
              …                   High portability
                                        …
Smart Vision Systems design =
Matching Application – Algorithm - Architecture

                   Application
                 (Requirements)




   Algorithm                        Architecture
  (Complexity)                     (Constraints)
Smart Vision Systems design =
Matching Application – Algorithm - Architecture

                   Application
                 (Requirements)




   Algorithm                        Architecture
  (Complexity)                     (Constraints)
Smart Vision Systems design =
Matching Application – Algorithm - Architecture

                   Application
                 (Requirements)




   Algorithm                                 Architecture
  (Complexity)     Introduce parallelism
                                            (Constraints)
                  Processor optimizations
Complexity - Quality trade-off: reduce complexity with
limited (visual) quality loss
Matching Algorithm and Architecture: the DCT example
f0                           F0


f1                           F1


f2                           F2


f3                           F3


f4                           F4


f5                           F5


f6                           F6


f7                           F7

        Classical DCT                 Butterfly DCT
        64 multiplications           20 multiplications
        64 additions                 26 additions

        High regularity              Low regularity

        Ideal for:                   Ideal for:
Smart Vision Systems design =
Matching Application – Algorithm - Architecture
                   Companies

                   Application
                 (Requirements)




                     IBBT




   Algorithm                        Architecture
  (Complexity)                     (Constraints)
Conclusions

Smart Vision Systems:
  Enable new applications
  Require new complex algorithms
  Use diverse platform architectures



Successful design requires competences on these 3
aspects

   IBBT brings these competences together
Demo’s
Demo 1: 3D video through real-time viewpoint
  interpolation
                              Viewpoint interpolation:
                              a)   stereo capturing
                              b)   depth extraction
                              c)   interpolation




Autostereoscopic displays:
       require multiple
     (interpolated) views
Demo 2: Eye-gaze corrected video chatting
                Webcam


                   Computer
                   Display
Demo 3: Novel 3D Camera prototype and monitoring
application in elderly environment

More Related Content

ODP
Alternativa3D_en
PPTX
IGARSS-SAR-Pritt.pptx
PPTX
Compressive Light Field Displays
PPTX
SIGGRAPH 2012 Computational Display Course - 3 Computational Light Field Disp...
PDF
iDiff 2008 conference #11 IP-Racine Barco
PDF
Build Your Own Multi-Touch Interface with Java and JavaFX Technology
PDF
Imagine Doc C
PDF
Discrete cosine transform
Alternativa3D_en
IGARSS-SAR-Pritt.pptx
Compressive Light Field Displays
SIGGRAPH 2012 Computational Display Course - 3 Computational Light Field Disp...
iDiff 2008 conference #11 IP-Racine Barco
Build Your Own Multi-Touch Interface with Java and JavaFX Technology
Imagine Doc C
Discrete cosine transform

What's hot (14)

PDF
Java me 08-mobile3d
PDF
Modern Graphics Pipeline Overview
PDF
Practical and Robust Stenciled Shadow Volumes for Hardware-Accelerated Rendering
PPTX
Keynote - SPIE Stereoscopic Displays & Applications 2014
PDF
Samsung box shc-735 e
PDF
Compressive DIsplays: SID Keynote by Ramesh Raskar
PDF
"High-resolution 3D Reconstruction on a Mobile Processor," a Presentation fro...
PDF
Keynote at 23rd International Display Workshop
PDF
Keynote Virtual Efficiency Congress 2012
PDF
NIAR_VRC_2010
PDF
PDF
Scv 3080 p
PPTX
Monocular simultaneous localization and generalized object mapping with undel...
PDF
Scd 2080 p
Java me 08-mobile3d
Modern Graphics Pipeline Overview
Practical and Robust Stenciled Shadow Volumes for Hardware-Accelerated Rendering
Keynote - SPIE Stereoscopic Displays & Applications 2014
Samsung box shc-735 e
Compressive DIsplays: SID Keynote by Ramesh Raskar
"High-resolution 3D Reconstruction on a Mobile Processor," a Presentation fro...
Keynote at 23rd International Display Workshop
Keynote Virtual Efficiency Congress 2012
NIAR_VRC_2010
Scv 3080 p
Monocular simultaneous localization and generalized object mapping with undel...
Scd 2080 p
Ad

Viewers also liked (20)

PDF
Trong tam ot tot nghiep
PDF
Apa2000 Ap Manual
PDF
Determinants of bank's interest margin in the aftermath of the crisis: the ef...
PPT
Trends And Drivers
PPTX
Benny Salaets - Realising the Digital Agend in Flanders
PDF
Complementary trust: IDEF Registry and Kantara cross-attestation
PDF
Oasis cloud-law-ics-unofficial
ODP
Moisés paisaxes-galego
PDF
Intangibles and Cultural and Creative Industries
PDF
Maduf01 Openings Speech Andre De Vleeschouwer
PDF
Q932+log reference fa lec 4 x1
PDF
4 deus leaflet wp4
PPT
Mobile National Days 2011 english
PDF
08 Afsluitevent Transecare
PDF
Open source eu-ict-ipr-clark-2010final
PDF
Ehip2 caring through sharing the ehip-project dirk colaert
PPT
GROUP5-SYLLABLES
PDF
La Contribución Socioeconómica de la Universidades Pública Valencianas. Terce...
PDF
OASIS at ITU/NGMN: Convergence, Collaboration and Smart Shopping in Open Stan...
PDF
I Minds2009 Lunch Event Icon I Minds 2009
Trong tam ot tot nghiep
Apa2000 Ap Manual
Determinants of bank's interest margin in the aftermath of the crisis: the ef...
Trends And Drivers
Benny Salaets - Realising the Digital Agend in Flanders
Complementary trust: IDEF Registry and Kantara cross-attestation
Oasis cloud-law-ics-unofficial
Moisés paisaxes-galego
Intangibles and Cultural and Creative Industries
Maduf01 Openings Speech Andre De Vleeschouwer
Q932+log reference fa lec 4 x1
4 deus leaflet wp4
Mobile National Days 2011 english
08 Afsluitevent Transecare
Open source eu-ict-ipr-clark-2010final
Ehip2 caring through sharing the ehip-project dirk colaert
GROUP5-SYLLABLES
La Contribución Socioeconómica de la Universidades Pública Valencianas. Terce...
OASIS at ITU/NGMN: Convergence, Collaboration and Smart Shopping in Open Stan...
I Minds2009 Lunch Event Icon I Minds 2009
Ad

Similar to 2008 brokerage 04 smart vision system [compatibility mode] (20)

PDF
“Tools for Creating Next-Gen Computer Vision Apps on Snapdragon,” a Presentat...
PPTX
PCI Geomatics Overview
PPTX
High-Level Synthesis Skill Development Needs - IEDEC
PDF
Scrambling For Video Surveillance
PPTX
IMAGE PROCESSING
PDF
Imaging automotive 2015 addfor v002
PDF
Imaging automotive 2015 addfor v002
PDF
Hardware Accelerated 2D Rendering for Android
PPT
Remote Graphical Rendering
PDF
How To Make Multi-Robots Formation Control System
PDF
Fujifilm - e2v L3 Vision
PDF
Training Videovigilancia IP: Capture
PDF
“Tensilica Processor Cores Enable Sensor Fusion for Robust Perception,” a Pre...
PPTX
Gentek Introduce(en)
PDF
Kinect v1+Processing workshot fabcafe_taipei
PPTX
System-on-Chip Programmable Retina
PDF
Discrete cosine transform
PDF
Discrete cosine transform
PPTX
Advanced Silverlight
PDF
“Tools for Creating Next-Gen Computer Vision Apps on Snapdragon,” a Presentat...
PCI Geomatics Overview
High-Level Synthesis Skill Development Needs - IEDEC
Scrambling For Video Surveillance
IMAGE PROCESSING
Imaging automotive 2015 addfor v002
Imaging automotive 2015 addfor v002
Hardware Accelerated 2D Rendering for Android
Remote Graphical Rendering
How To Make Multi-Robots Formation Control System
Fujifilm - e2v L3 Vision
Training Videovigilancia IP: Capture
“Tensilica Processor Cores Enable Sensor Fusion for Robust Perception,” a Pre...
Gentek Introduce(en)
Kinect v1+Processing workshot fabcafe_taipei
System-on-Chip Programmable Retina
Discrete cosine transform
Discrete cosine transform
Advanced Silverlight

More from imec.archive (20)

PDF
iMinds-iLab.o, Open Innovation in ICT
PDF
Accio presentation closing event
PPTX
PRoF+ Patient Room of the Future
PPTX
Results of the Apollon pilot in homecare and independent living
PPTX
Delivery of feedback on Health, Home Security and Home Energy in Aware Homes ...
PDF
NMMU-Emmanuel Haven Living Lab
PDF
The Humanicité workshops
PPTX
A Real-World Experimentation Platform
PDF
ENoLL @ AAL Forum 2012
PDF
ENoLL 6th Wave Results Ceremony (Jesse Marsh)
PDF
The Connected Smart Cities Network and Living Labs - Towards Horizon 2020 - K...
PDF
Apollon-23/05/2012-9u30- Parallell session: Living Labs added value
PPT
Apollon - 22/5/12 - 11:30 - Local SME's - Innovating Across borders
PPT
Apollon - 22/5/12 - 16:00 - Smart Open Cities and the Future Internet
PPT
Apollon - 22/5/12 - 16:00 - Smart Open Cities and the Future Internet
PPT
Apollon - 22/5/12 - 16:00 - Smart Open Cities and the Future Internet
PPTX
Apollon - 22/5/12 - 16:00 - Smart Open Cities and the Future Internet
PPTX
Apollon - 22/5/12 - 11:30 - Local SME's - Innovating Across borders
PPTX
Apollon - 22/5/12 - 09:00 - User-driven Open Innovation Ecosystems
PPT
Apollon - 22/5/12 - 09:00 - User-driven Open Innovation Ecosystems
iMinds-iLab.o, Open Innovation in ICT
Accio presentation closing event
PRoF+ Patient Room of the Future
Results of the Apollon pilot in homecare and independent living
Delivery of feedback on Health, Home Security and Home Energy in Aware Homes ...
NMMU-Emmanuel Haven Living Lab
The Humanicité workshops
A Real-World Experimentation Platform
ENoLL @ AAL Forum 2012
ENoLL 6th Wave Results Ceremony (Jesse Marsh)
The Connected Smart Cities Network and Living Labs - Towards Horizon 2020 - K...
Apollon-23/05/2012-9u30- Parallell session: Living Labs added value
Apollon - 22/5/12 - 11:30 - Local SME's - Innovating Across borders
Apollon - 22/5/12 - 16:00 - Smart Open Cities and the Future Internet
Apollon - 22/5/12 - 16:00 - Smart Open Cities and the Future Internet
Apollon - 22/5/12 - 16:00 - Smart Open Cities and the Future Internet
Apollon - 22/5/12 - 16:00 - Smart Open Cities and the Future Internet
Apollon - 22/5/12 - 11:30 - Local SME's - Innovating Across borders
Apollon - 22/5/12 - 09:00 - User-driven Open Innovation Ecosystems
Apollon - 22/5/12 - 09:00 - User-driven Open Innovation Ecosystems

2008 brokerage 04 smart vision system [compatibility mode]

  • 1. Smart Vision System: Design of Application Algorithm-Architecture Eric Delfosse IBBT-NES-IMEC
  • 2. What are Smart Vision Systems? Systems that embed intelligence through advanced image processing to: Enhance visual user experience Improve natural interaction Facilitate decision making for complex events …
  • 3. Enhance visual user experience Multi-camera 3D image reconstruction for advanced surveillance 3D GPS (IBBT GBO ISYSS) (IBBT GBO URBAN)
  • 4. Increase natural interaction Natural human-machine interface through gesture recognition 3D (immersive) video (IBBT GBO Hi-Masquerade) conferencing
  • 5. Facilitate decision making for complex events Event detection for surveillance applications (IBBT GBO ISYSS) Traffic sign recognition for driver assistance (IBBT GBO URBAN)
  • 6. Different applications, different requirements Throughput High resolution High framerate Functionality Accuracy Reduce false negatives and false positives … Size Low Power Solar energy Avoid active cooling ... Cost
  • 7. These applications require increasingly complex algorithms Exponential algorithmic complexity increase > 9000 3000 2500 Complexity (Ops/pixel) 2000 1500 1000 500 0 enhancement extraction Depth extraction Edge detection AVC encoding Object detection understanding Feature Image Image Algorithm
  • 8. Diversity of platform architectures with different characteristics PC + GPU High performance High power Limited portability (laptop) … Server rack Embedded systems Very high performance Very high Power Medium performance Non-portable Low power … High portability …
  • 9. Smart Vision Systems design = Matching Application – Algorithm - Architecture Application (Requirements) Algorithm Architecture (Complexity) (Constraints)
  • 10. Smart Vision Systems design = Matching Application – Algorithm - Architecture Application (Requirements) Algorithm Architecture (Complexity) (Constraints)
  • 11. Smart Vision Systems design = Matching Application – Algorithm - Architecture Application (Requirements) Algorithm Architecture (Complexity) Introduce parallelism (Constraints) Processor optimizations
  • 12. Complexity - Quality trade-off: reduce complexity with limited (visual) quality loss
  • 13. Matching Algorithm and Architecture: the DCT example f0 F0 f1 F1 f2 F2 f3 F3 f4 F4 f5 F5 f6 F6 f7 F7 Classical DCT Butterfly DCT 64 multiplications 20 multiplications 64 additions 26 additions High regularity Low regularity Ideal for: Ideal for:
  • 14. Smart Vision Systems design = Matching Application – Algorithm - Architecture Companies Application (Requirements) IBBT Algorithm Architecture (Complexity) (Constraints)
  • 15. Conclusions Smart Vision Systems: Enable new applications Require new complex algorithms Use diverse platform architectures Successful design requires competences on these 3 aspects IBBT brings these competences together
  • 17. Demo 1: 3D video through real-time viewpoint interpolation Viewpoint interpolation: a) stereo capturing b) depth extraction c) interpolation Autostereoscopic displays: require multiple (interpolated) views
  • 18. Demo 2: Eye-gaze corrected video chatting Webcam Computer Display
  • 19. Demo 3: Novel 3D Camera prototype and monitoring application in elderly environment