| Summary: | The complexity of implementing vision algorithm on embedded systems can greatly benefit from research in HW/SW partitioning and IP-reuse. This paper presents a novel research work of a hybrid HW/SW partitioning method that combines heuristic and knowledge-based approaches to satisfy user-defined constraints. In order to achieve this objective, Bayesian Belief Network (BBN) is utilised and incorporated into the framework to produce a reliable HW/SW partitioning for a given vision algorithm. To provide a better convergence, software weight is incorporated into the link matrices. The outcome of the framework will be the partitioned modules that satisfy the user-defined timing and resource constraints. In this paper, we also report on comparison of our proposed framework with the previous work reported in the literature including: BBN by University of Arizona, the exhaustive algorithm and the greedy algorithm.
|
|---|