2008 brokerage 04 smart vision system [compatibility mode]
0 likes197 views
Smart vision systems embed intelligence through advanced image processing to enhance visual experiences, improve interactions, and facilitate decisions. They require increasingly complex algorithms to perform tasks like 3D reconstruction, gesture recognition, and event detection. These algorithms must be matched to diverse platform architectures with different performance and power constraints. Successful system design requires matching applications, algorithms, and architectures. The document discusses examples including 3D video interpolation, eye-gaze corrected video chatting, and a 3D camera prototype for elderly monitoring. It concludes that IBBT brings together competences in applications, algorithms, and architectures to enable new smart vision systems.
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
- 16. Demo’s
- 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