1. Smart Lab - RBV Network 1 Radial Basis Voronoi Network: An Internet Enabled Vision System for Remote Object Classification RAYMOND K. CHAFIN, CIHAN H. DAGLI Smart Engineering Systems Laboratory, University of Missouri – Rolla, Rolla, MO 65409-1060 chafin@advertisnet.com, dagli@umr.edu RAYMOND K. CHAFIN, CIHAN H. DAGLI Smart Engineering Systems Laboratory, University of Missouri – Rolla, Rolla, MO 65409-1060 chafin@advertisnet.com, dagli@umr.edu

2. Smart Lab - RBV Network 2 Project Goals –Demonstrate innovative neural network architecture for 3D object classification –Provide network communications for distributed client-server architectures –Visibility into all levels of processing –Handle variable image frame sizes and formats –Rapid reconfiguration and extension –User-friendly graphical interface –Demonstrate innovative neural network architecture for 3D object classification –Provide network communications for distributed client-server architectures –Visibility into all levels of processing –Handle variable image frame sizes and formats –Rapid reconfiguration and extension –User-friendly graphical interface

3. Smart Lab - RBV Network 3 RBV Network Processing Steps Image Acquisition Edge Detection Edge Enhancement Edge Skeletonization Vertex Extraction Vertex Connection and Web Build Perceptual Grouping and Segmentation Invariant Transform Object Learning Object Recognition HIGHER VISION INTERMEDIATE VISION EARLY VISION

4. Smart Lab - RBV Network 4 Performance Considerations Effective classification technique for 3D objects in arbitrary posesEffective classification technique for 3D objects in arbitrary poses Multi-object recognition in single image framesMulti-object recognition in single image frames Clutter removalClutter removal Computational efficiencyComputational efficiency Internet communications protocols minimize bandwidth and simplify handshakingInternet communications protocols minimize bandwidth and simplify handshaking Effective classification technique for 3D objects in arbitrary posesEffective classification technique for 3D objects in arbitrary poses Multi-object recognition in single image framesMulti-object recognition in single image frames Clutter removalClutter removal Computational efficiencyComputational efficiency Internet communications protocols minimize bandwidth and simplify handshakingInternet communications protocols minimize bandwidth and simplify handshaking

5. Smart Lab - RBV Network 5 Coding Considerations Accessibility to studentsAccessibility to students Modular “toolbox” architectureModular “toolbox” architecture Object-oriented codingObject-oriented coding Optimized computational efficiencyOptimized computational efficiency MATLAB best suited to UMR studentsMATLAB best suited to UMR students C++ efficient, object-orientedC++ efficient, object-oriented Hybrid architecture favoredHybrid architecture favored Accessibility to studentsAccessibility to students Modular “toolbox” architectureModular “toolbox” architecture Object-oriented codingObject-oriented coding Optimized computational efficiencyOptimized computational efficiency MATLAB best suited to UMR studentsMATLAB best suited to UMR students C++ efficient, object-orientedC++ efficient, object-oriented Hybrid architecture favoredHybrid architecture favored

6. Smart Lab - RBV Network 6 Mixed-language Architecture Matlab Graphical User Interface Matlab Graphical User Interface Single Function Gateway Routine (Win32 DLL )... 84 Total Gateway Functions... 84 Total Gateway Functions RBV NET Main Engine (Win32 DLL) Multiple functions coded in C++ Copies reside on both client and server RBV NET Main Engine (Win32 DLL) Multiple functions coded in C++ Copies reside on both client and server Internet Client- Server Modules Camera Interface Modules

7. Smart Lab - RBV Network 7 Graphical User Interface

8. Smart Lab - RBV Network 8 Related Studies L1 and L2 Primate Retinal Models Human Lateral Geniculate Nucleus and V1 Cortex segment completion models Adaptive linear image filtering Genetic algorithms Principal Component Analysis / Independent Component Analysis L1 and L2 Primate Retinal Models Human Lateral Geniculate Nucleus and V1 Cortex segment completion models Adaptive linear image filtering Genetic algorithms Principal Component Analysis / Independent Component Analysis

9. Smart Lab - RBV Network 9 Results To Date Realizing exceptional feature extraction accuracy versus speed Highly accurate classifications on a variety of 3D object types, including human faces Detailed behavior of network processing steps made visible Modular architecture facilitates rapid implementation of new algorithms Code library at 50,000 lines Realizing exceptional feature extraction accuracy versus speed Highly accurate classifications on a variety of 3D object types, including human faces Detailed behavior of network processing steps made visible Modular architecture facilitates rapid implementation of new algorithms Code library at 50,000 lines

10. Smart Lab - RBV Network 10 Future Work Improved feature extraction - oriented hysteresisImproved feature extraction - oriented hysteresis Object isolation using techniques developed for MPEG video compressionObject isolation using techniques developed for MPEG video compression Similarity metrics for large object setsSimilarity metrics for large object sets Adaptive vigilanceAdaptive vigilance Improved feature extraction - oriented hysteresisImproved feature extraction - oriented hysteresis Object isolation using techniques developed for MPEG video compressionObject isolation using techniques developed for MPEG video compression Similarity metrics for large object setsSimilarity metrics for large object sets Adaptive vigilanceAdaptive vigilance