10
the specific problems [22, 24]. CNNs are a special type of ANN consisting of a grid of cells
which are connected to eachother locally. The local connectivity makes CNNs more suitable
for hardware implementation (while compared to other ANN which require a global
connectivity) [25]. The main advantage of CNN is that by changing two templates matrices
one can change the functionality of the CNN processor without any hardware
reconfiguration. CNN cells do work in parallel and therefore ensure an ultrafast
manipulation of pixels [26].
For processing information based on neuro-computing we have to consider parallel
computation, learning method and adaptation [27]. Depending on the specific task, we can
define a learning rule for training the network [28]. This method can play an important
role for solving complex and time consuming problems in machine vision. Pattern
recognition, optimization, classification and image enhancements are important tasks in
ADAS concept for which we can use neuro-computing techniques [28]. In ADAS concepts,
the key issues are real-time processing and robustness. Neuro-computing can provide very
stable, accurate and ultra fast solutions for various problems. In most cases, the training
phase is slow and time consuming but at the end, that is after the training phase, operating
artificial neural networks it is very fast [24].
