Detect which regions in an image contain motion. Currently hardwired to nine regions (1-9), corresponding to upper-left, upper-middle, upper right, etc:

patch