Download presentation
Presentation is loading. Please wait.
Published byMeagan Griffin Modified over 9 years ago
1
Hierarchical model for pattern recognition based on parallel and distributed computing Olivier Bornet, University Joseph Fourier Grenoble Martin Kalany, Vienna University of Technology Scientific advisor: Sergey V. Axyonov
2
Motivation and targets Motivation Object recognition is a widely used feature from industrial robots to autonomous vehicles Multiple-Core processors are now fairly common Image proccessing is highly depending on processor- intensive operations with matrices Targets Create and implement a model that is capable of recognizing simple features Use concurrent and distributed programming techniques to massivly decrease execution time 2 Olivier Bornet, Martin Kalany: Hierarchical model for pattern recognition based on parallel and distributed computing
3
Overview of the model 3 Olivier Bornet, Martin Kalany: Hierarchical model for pattern recognition based on parallel and distributed computing
4
Mathematical model Filter layer – Detects rather small features like lines with different angles and features – Uses different Gabor filters for features detection 4 Olivier Bornet, Martin Kalany: Hierarchical model for pattern recognition based on parallel and distributed computing -Feature angle -Feature width -Pixel location inside of cell receptive field - Output of filter cell with location - Gray-scale input - Receptive field
5
Mathematical model MAX layer – Combination of the results of several different Gabor filters – Detect local features regardless of it‘s width 5 Olivier Bornet, Martin Kalany: Hierarchical model for pattern recognition based on parallel and distributed computing - Output of MAX cell with location - Plane index Kohonen layer – Stores information about more complex features - Estimates Euclidian distance between MAX layer outcomes and stored features
6
Mathematical model 6 Olivier Bornet, Martin Kalany: Hierarchical model for pattern recognition based on parallel and distributed computing View-tuned units (VTU) - Remembers all complex features of entire object -Uses radial-basis functions to estimate similarity measure between input object and stored patterns VTU output: - Kohonen layer outputs
7
Algorithm 7 Olivier Bornet, Martin Kalany: Hierarchical model for pattern recognition based on parallel and distributed computing
8
OpenMP &MPI 8 Olivier Bornet, Martin Kalany: Hierarchical model for pattern recognition based on parallel and distributed computing
9
Results 9 Olivier Bornet, Martin Kalany: Hierarchical model for pattern recognition based on parallel and distributed computing Constructed model All layers successfully implemented Studied principles and usage of parallel and distributed programming Parallel programming already in use for Kohonen layer
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.