Julian Eggert, Volker Willert, Jens Schmüdderich, "Biological Motion Estimation with Spatiotemporal Integration", HRI Global Workshop, 2007.

We present a framework and study a model for biologically inspired motion estimation. It comprises several “areas” (corresponding roughly to V1-MT) which locally analyze motion input at different levels of granularity, coupled to functionally distinct areas which concentrate on the extraction of more global motion patterns (corresponding e.g. to MST). The areas are coupled reciprocally with each other, with “higher” areas providing prior knowledge or constraints that bias the processing at lower areas. In addition, within each area, spatial and temporal integration, mediated by lateral connections and internal dynamics, is used to resolve the inherent ambiguity of motion signals and to take advantage of coherence in naturally occurring stimuli. Our model leads to results that are quantitatively comparable to state-of-theart techniques. We show how several psychophysical as well as physiological effects observed during motion processing can be explained by our model. We also show implementation results of our model in the application contexts of motion segmentation and egomotion compensation for ASIMO.