I have been using TPL dataflow for an image processing pipeline using the producer/consumer pattern.
I'm trying to work out the best approach to allow for algorithms that require either the previous frame or a persistent object.
An example of one of these processes is background subtraction where the foreground mask is found by comparing the current frame to a dynamic background model.
One idea of how to achieve this is to create a cycle in the pipeline:
Is there a standard way of approaching this sort of pipeline?
Are there any issues related to the ordering of data or the asynchronous operation?