Fig 4 shows the data flow of the whole system. The host PC
performs the classification and abstraction of the vision data
and this data logs important parameters to be compared with
the embedded target at hand. The algorithms we discussed
support the construction of a road model, calculation of the
vehicle width and classification of pedestrian and traffic sign
target instances. The video is replayed onto the embedded
target and checked for its ability to calculate the information
shown in the Fig 4. The information calculated over the time
stamps is measured and compared with information calculated
on the host PC. Performance characterization results based on
the parameters discussed can be easily automated with this
data flow and algorithm analysis becomes a much less tedious
process.