116
then propose a new concept that can satisfy all the
cited requirements in ADAS
systems including flexibility in design.
Research question 2: What are the major limitations of traditional high
performance computing approaches if used to ensure “rea
l-
time” image
processing in ADAS ?
For this research question we studied about high performance computing and real-
time image processing. Also, we could show the limitation of
traditional computing
concepts/architectures based on the
Von Neumann
architecture.
We have shown
that manipulating and processing pixels in parallel does speed the image
processing. For flexibility in design, hardware should be reconfigurabe by software
in run-time mode and without any need for reprogramming the system.
Research question 3: What is the huge potential of neurocomputing involving
either traditional neural networks (NN) or cellular neural networks (CNN)
for high-speed and flexible image processing for ADAS? Are there any
limitations and how can these eventually be addressed?
For this question we have shown that ANN has
huge potential for image
processing; examples of applications are pattern recognition,
feature extraction,
compression, etc. Also we have mentioned some drawbacks of ANN for hardware
implementation and a comparison between CNN and ANN. Overall neurocomputing
is a paradigm that promises to solve the tough requirements of ADAS concerning
computing speed and flexibility.
