Parallel implementation and evaluation of motion estimation systemalgorithms on a distributed memory multiprocessor using knowledge basedmappings

摘要

Several techniques to perform static and dynamic load balancingfor vision systems are presented. These techniques capture thecomputational requirements of a task by examining the data when it isproduced. They can be applied to many vision systems because manyalgorithms in different systems are either the same or have similarcomputational characteristics. These techniques are evaluated byapplying them on a parallel implementation of the algorithms in a motionestimation system on a hypercube multiprocessor system. It is shown thatthe performance gains when these data decomposition and load balancingtechniques are used are significant and that the overhead of using thesetechniques is minimal