Tsotsos et al. 1984 : The goal of this paper was to briefly discuss the relationship between low and high levels of visual processing in the context of the real-time interpretation of a dynamic environment. A central claim was that the consideration of time cannot be an 'add-on' to a strictly spatial interpretation scheme. In particular, for real-time motion understanding systems, several major points regarding time have been discussed. These are: finite computation time while the environment continues to change; finite storage space necessitating temporal abstraction; and stability with noise which requires temporal integration. The distinctions between high and low level computations were highlighted and an intrinsic image-like representation proposed for their interface. In addition, the terms non-attentive and attentive vision were introduced and their psychological relevance discussed. A computational framework for motion understanding systems was described that has its roots in a previous framework developed at the University of Toronto. This framework involves both high and low level components, both incorporating organization hierarchies for efficiency and abstraction.