This component can be used to find moving foreground features, which can be a powerful way to suppress false background detections in subsequent tracking steps.
set time window, and standard deviations above background for foreground
time window should be more than 2x larger than time taken for a feature to traverse a pixel (NB. total window is 2x half-width +1)
moving foreground identified by intensity increase relative to background average (i.e. median) for a pixel over a given time window
"soft" segmentation, yielding foreground probability related to excess intensity (in standard deviations) over background level
crude Anscombe transform applied to data to stabilize the variance