Exploiting Parallelism in Multi-View Systems Using UML Activity Diagrams and OpenMP

Many demanding Pattern Recognition tasks are now possible in embedded systems, since multiprocessor designs are becoming a common execution platform in this area. The use of application-level concurrency allows to increase system throughput and to reduce the execution time of embedded applications in such systems. The UML standard provides mechanisms to model the concurrent control and data flows of an application through its activity diagrams. However, manual transformation of these models into high-level languages with concurrency support is an error-prone process. Model-Driven Engineering approach advocates for an automatic process to translate models into high-level programming languages. This work deals with the transformation of UML activity diagrams representing data and control flows of a multi-view PR application into parallel programs based on the OpenMP standard. The resulting application code not only is highly portable but it also allows to easily control the application concurrency level to customize the code for a specific embedded platform.