This paper presents a framework for computer vision tasks on Field Programmable Gate Arrays (FPGA) which allows rapid integration of vision algorithms by separating the framework from the vision algorithms. A vision system can be created by using plug-and-play methodology. On an abstract level several input and output channels of the system can be defined. Also, commonly used image transformations are modularized and can be added to the inputs or outputs of an algorithm. Special input and output modules allow the integration of algorithms with no knowledge of the surrounding framework.
Image processing framework for fpgas: Introducing a plug-and-play computer vision framework for fast integration of algorithms in reconfigurable hardware.