Research Project

DARPA Unconventional Processing of Signal for Intelligent Data Exploitation (UPSIDE)

Designing a complete MS CMOS-based system for image pre-processing, segmentation, tracking, and classification.

With the volume of visual sensor data increasing exponentially, there is a dramatic increase in the complexity of analysis, reflected in the number of operation per pixel per second.  This project involves designing a complete MS CMOS-based system for image pre-processing, segmentation, tracking, and classification.  In doing so, we utilize the IFAT (Integrate and Fire Transceiver) model and ideas for computing various image processing functions.  It is a mixed-signal VLSI model of a capacitor array representing an array of I&F neurons which integrate and spike using AER communication to send/receive spikes/events.  The idea of using this biologically-plausible analog-VLSI approach for visual processing is ideal for processing speed and power efficiency.  Furthermore, we apply unconventional Bayesian Approximation and Inference algorithms which use stochastic computation to also allow for a large reduction in amount of data required for accurate computations.  We can further apply these ideas to numerous other vision-related applications.

Back to top