Key Issues
CAVIAR is intended to provide a framework for the implementation of artificial
vision systems and other bioinspired processing systems that have to be
compact and work in real time. In summary, the benefits of the proposal
are:
1. To provide a robust,
user-friendly infrastructure that will enable non-experts to assemble hierarchically
structured multi-layer multi-chip neural sensing/processing/actuation systems.
This infrastructure will be based on the emerging AER technique.
2. To provide a novel AER-based technique to perform two-dimensional convolutions
at high speeds, which is a very desirable feature for artificial vision
systems that aim to work in real time.
3 . To provide a medium-resolution sensing retina that extracts contrast
information from an image and is
insensitive to lighting conditions.
4 . To develop high-level processing chips for dimension reduction, competition
and the learning of spatiotemporal patterns.
5. To provide a small demonstrator that makes use of the developed building
blocks. This demonstrator will be a proof of concept and a stepping stone
for long-term research on AER-based bio-inspired neural systems. It will
also be used to present the benefits of our approach to industries and research
institutes with an interest in compact realtime data processing systems
for application fields in intelligent transportation (such as autonomous
car navigation), robotics, aeronautics, assistance to the impaired, prosthetics,
etc.
6. To make the emerging AER technique accessible to non-engineers in research
and industry that are interested in developing and using bio-inspired hierarchically
structured neural perception and processing systems.
