Qualitative Motion

The analysis and modelling of processes involved in motion perception is the aim of our interdisciplinary cooperation between psychophysics and computer science 3. This paper proposes a qualitative representation of the course of motion that was the outcome of the computer science part of the project. A course of motion is described at two diierent layers that provide diierent levels of granularity, abstraction and accuracy. A vectorial representation at the lower level is qualitative, but still quite accurate and allows switches between deictic and intrinsic frames of reference. A propositional representation at the higher level is more abstract, less accurate and has coarser granularity, but reeects something of the \semantics" of the course of motion, in that it identiies meaningful subsequences in the vectorial representation and sums them up to a higher structure. The propositional layer is constructed from the vectorial layer by segmentation of the vector sequence and classiication of the subsequences according to their shape.