M. Langer and R. Mann, Dimensional Analysis of Image Motion, Eighth International Conference on Computer Vision. Vancouver, Canada. I:155--162. July 2001.
Abstract: Previous studies of image motion have addressed particular types of motion on a case-by-case basis: a small moving object, a moving contour, a 2D optical flow field, motion across an occluding boundary. Here we introduce a categorical framework for image motion that supports the categories addressed in previous studies and that relates these known categories to each other and to several new categories that have not been considered previously. The categories are expressed in terms of the dimensionality of the motion. We consider a specific new category in detail. The category occurs when an observer moves relative to a massively cluttered scene. Examples are observer motion through a forest or crowd, and passive viewing of falling snow. Classical optical flow techniques are inappropriate for analyzing such motions since they assume a unique motion at each pixel. We introduce a new method for analyzing such motions and show results for synthetic and real image sequences.