CAMPOS, N. C. S.; http://lattes.cnpq.br/3560052420335314; CAMPOS, Nelson Carlos de Sousa.
Resumo:
A methodology for hardware design and validation of face detection systems is
discussed. The design begins at a high-level of abstraction by using a software library as
the golden reference model. The golden reference model and the baseline model (model
under validation) are converted to a transaction level (TLM) description or a register
transfer level (RTL) description. The degree of similarity of the detected faces is used to
determine whether the model under validation fulfills the prescribed design requirements.
The methodology is conceived for real time high resolution image processing applications
and is based on the universal verification methodology (UVM). As a case study, the
methodology was applied to validate a C++ baseline model written to implement the
OpenCV library considered as the golden reference model. In this case both models have
been converted to TLM and thus the time required for determining whether the baseline
model is valid or not is significantly reduced. Thus, one can decide before proceeding with
the conversion of the baseline model to RTL, if the TLM model needs improvement.