7
not trivial to speed up them by the using pipelining or parallel
processing and it needs
different model of processing [13].
1.2
Research questions and objectives of the thesis
In the following we do briefly describe and justify the key research questions of this thesis
as well as the core of answers/solutions that have been obtained for each of them. Overall,
we have formulated seven key questions concerning ADAS technology, importance of high
performance computing and hardware implementation.
Research question 1: What are the hard requirements of ADAS concerning real-time
image processing and design flexibility? How far do traditional
approaches fail to satisfy these requirements?
One of the main issues to design an ADAS system is high processing speed and robustness
of the environment perception (image processing based).
The maximum interval for
processing frames should not be greater than 20ms. In some cases such as lane departure
warning and collision detection this level changes to maximum 10 ms processing time [14,
15]. The fundamental issue in all image processing systems is robustness and accuracy. In
ADAS technology which is based on image processing and machine vision we have to
guarantee the stability
and robustness of detection, identification and recognition of
features [16]. For example a lane departure warning
system should work in any
environmental condition and lighting [16, 17].
Concerning video- based ADAS solutions different image processing filters and appropriate
hardware architectures are required. Some filters are very complex and very demanding in
terms of processing run time. Hence, we are interested in heavy parallel processing and
task concurrency. In most of different ADAS concepts we one is using the same processing
modules with different connectivity and topology. Hence, if we have a reconfigurable
architecture or a universal model we do not need to reserve hardware resources for each
different concept.
While following traditional methodological ways of doing and design one does use/involve
sequential
processing
architecture,
time
sharing
and
multi-threading
8
algorithms/processing. The weakness of this traditional way of doing concerning
processing performance is that it results in a ‘too long’ processing
time and therefore
making it very difficult to satisfy real-time constraints. The real-time constraints expect a
completion of the processing of a frame within a time window that is less than the
capturing time of that frame. Therefore, for a 60 FPS (frames per second) rate we do have a
maximum of 15 milliseconds to finish all the processing. And since one does generally need
about 6~10 different high definition (HD) image preprocessing modules whereby each of
them takes around 5ms per frame we thus do reach a total ranging between 30 ms and
50ms of processing time if one tries to use the traditional processing schemes. This figure
of 50 ms does fail to fulfill the real-
time constraint of “maximum 15 ms”
processing time
per frame. It is therefore clear that traditional processing schemes are not capable of
fulfilling the real-time constraints of visual sensors in/for ADAS..
Examples of ADAS
solution of relevance for visual sensors are: the
Lane Departure Warning
(LDW),
Adaptive
Cruise Control
(ACC), Emergency
Brake Assistant
(EBA) and
Blind Spot Detection
(BSD), etc.
The different ADAS solutions should be able to re-use the same components/modules for
the image processing. Another weakness of classical algorithms they do not enable an easy
re-use of functional components. Therefore we do need and are looking for a special
architecture that
is reconfigurable by software; such a concept will enable an easy re-use of
the same platform for different functionalities and algorithms.
