Co-design by parallel prototyping: optical-flow detection case study

Current OF routines have achieved greater accuracy over larger speeds by improvements in implementation. Unfortunately, the improvements require the processing of more image frames or larger spatial regions. General-purpose (gp) multicomputers can be deployed to reduce the timings recorded on workstations. The four methods parallelised herein are not unusual amongst OF methods in being amenable to data-farming. Parallelising a number of routines in a systematic manner is possible if a generalised framework is available such as is provided by the PPF methodology and knowledge of the common structure to OF methods. The advantage of parallel pipelines may be that independent stages of the pipeline can at a later time be transferred to more suitable hardware.