You can also compute optical flow on the depth image...
check viim sdk (
viim.pt), we compute optical flow vectors on the depth image (works with no light and is tested on 320x240).
Nevertheless, if one don't need the motion vectors, only the location where changes occur, frame differencing seems a good possibility.
ciao,
pedro
CoVii - Computer Vision Interaction