The First International Workshop on Computational Models of the Visual Cortex: Hierarchies, Layers, Sparsity, Saliency and Attention

Research Article

Event-based optical flow on neuromorphic hardware

  • @INPROCEEDINGS{10.4108/eai.3-12-2015.2262447,
        author={Tobias Brosch and Heiko Neumann},
        title={Event-based optical flow on neuromorphic hardware},
        proceedings={The First International Workshop on Computational Models of the Visual Cortex: Hierarchies, Layers, Sparsity, Saliency and Attention},
        publisher={ACM},
        proceedings_a={CMVC},
        year={2016},
        month={5},
        keywords={optical flow event-based sensing address-event representation real-time vision neuromorphic computing},
        doi={10.4108/eai.3-12-2015.2262447}
    }
    
  • Tobias Brosch
    Heiko Neumann
    Year: 2016
    Event-based optical flow on neuromorphic hardware
    CMVC
    ACM
    DOI: 10.4108/eai.3-12-2015.2262447
Tobias Brosch1,*, Heiko Neumann1
  • 1: Institute of Neural Information Processing | Ulm University
*Contact email: tobias.brosch@uni-ulm.de

Abstract

Event-based sensing, i.e. the asynchronous detection of luminance changes, promises low-energy, high dynamic range, and sparse sensing. This stands in contrast to whole image frame-wise acquisition using standard cameras. Recently, we proposed a novel biologically inspired efficient motion detector for such event-based input streams and demonstrated how a canonical neural circuit can improve such representations using normalization and feedback. In this contribution, we suggest how such a motion detection scheme is defined by utilizing a canonical neural circuit corresponding to the resolution of cortical columns. In addition, we develop a mapping of key computational elements of this circuit model onto neuromorphic hardware. In particular, we focus on the recently developed TrueNorth chip architecture by IBM to realize a real-time, energy-efficient and adjustable neuromorphic optical flow detector. We demonstrate the function of the computations of the canonical model and its approximate neuromorphic realization.