FPGA-accelerator system for computing biologically inspired feature extraction models

Neuromorphic algorithms for computer-based vision may be the next step towards improving the way computers gather and interpret visual information. However, these algorithms typically have high computational demands making them difficult to deploy in embedded environments where power consumption is equally as important as performance. In this paper, we present an embedded implementation of a ventral visual pathway model, HMAX. We describe an embedded FPGA system that implements the model, as well as accelerator engines necessary to ensure adequate performance. The final system is shown to operate within a power budget of 3W while achieving up to 16.5X speedup over a pure embedded processor implementation.

[1]  Thomas Serre,et al.  A Theory of Object Recognition: Computations and Circuits in the Feedforward Path of the Ventral Stream in Primate Visual Cortex , 2005 .

[2]  Christof Koch,et al.  A Model of Saliency-Based Visual Attention for Rapid Scene Analysis , 2009 .

[3]  Thomas Serre,et al.  Robust Object Recognition with Cortex-Like Mechanisms , 2007, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[4]  Tomaso Poggio,et al.  CNS: a GPU-based framework for simulating cortically-organized networks , 2010 .

[5]  David G. Lowe,et al.  University of British Columbia. , 1945, Canadian Medical Association journal.

[6]  T. Poggio,et al.  Regularized Least-Squares Classification 133 In practice , although , 2007 .

[7]  Berin Martini,et al.  Hardware accelerated convolutional neural networks for synthetic vision systems , 2010, Proceedings of 2010 IEEE International Symposium on Circuits and Systems.

[8]  Fei-FeiLi,et al.  Learning generative visual models from few training examples , 2007 .

[9]  Gustavo Deco,et al.  Computational neuroscience of vision , 2002 .

[10]  Narayanan Vijaykrishnan,et al.  A hardware architecture for accelerating neuromorphic vision algorithms , 2011, 2011 IEEE Workshop on Signal Processing Systems (SiPS).

[11]  Pietro Perona,et al.  Learning Generative Visual Models from Few Training Examples: An Incremental Bayesian Approach Tested on 101 Object Categories , 2004, 2004 Conference on Computer Vision and Pattern Recognition Workshop.

[12]  Jing Peng,et al.  SVM vs regularized least squares classification , 2004, ICPR 2004.

[13]  T. Poggio,et al.  Hierarchical models of object recognition in cortex , 1999, Nature Neuroscience.