196 CHAPTER 5. NEURAL NETWORKS FOR CONTROL
imaging the street-crossing signal. A real-time neural network processor
would then identify the signal/symbol and trigger a voice command that
allows the blind to either ìstopî or ìwalkî at the pedestrian crossing. We
are required to develop a neural network that can be trained to recognize
ìWALKî and ìSTOPî signals commonly found at pedestrian crossings.
These signals can be either in the form of words as indicated or symbols
such as a ìpersonî or a ìhand.î You are required to develop the appropri-
ate set of training patterns and train a neural network with four outputs.
Each of the outputs should represent the word or symbol used in training.
Discuss in detail the technical merits and practical issues in implementing
such a system.
13.Project: Color is a significant feature in several applications. In the
food industry, for example, color is used as an indicator of whether or not
thefoodisofgoodquality. Thereisadifference in color between fresh
vegetables and not-so-fresh vegetables. The same can be said about meat
and other food products. Color is also used in distinguishing between
objects such as lemons and lime. As we can see, the possibilities of using
color for the purpose of identification of objects are enormous and can be
explored in many different applications. We leave the choice of selecting
a project to the explorative and inquisitive mind.
As a motivation to pursue studies in neural networks, in this project how-
ever, you are required to develop a neural network that can distinguish
between the three colors red, yellow, and green that we commonly see at
traffic intersections. Not only is color important, we also recognize the
need to identify the direction indicated by the traffic light in terms of the
ìLeft Arrow,î the ìRight Arrow,î the ìUp Arrow,î and the ìDown Ar-
row.î Design a neural network that can identify traffic light conditions.
Discuss how such a system may be implemented. Identify and describe
specific technical issues that would have to be addressed if such a system
were implemented.
14.Project: Neural network-based applications are useful wherever there
is a need to bridge the gap in communication between humans and ma-
chines. Brain-machine interfaces is a new area of scientificexploration
where the objective is to make machines respond to human voice, ges-
tures, and thinking. One area that such an application is significant, and
can improve the communication ability, is to provide an interface that can
translate computer-generated commands to Braille and vice versa. This
would be a significant aid to the blind. In this project, we ask that a set
of training patterns be selected thatadequately represent commonly used
words and commands. For example, in using a word processor we choose
the ìFile,î ìEdit,î ìSave,î and several other menu items to performfile
management operations. It would be important to translate these into
Braille so that a blind person can use the word processor effectively. Al-
ternatively, if the computer response were ìDo you wish to replace an
