Probabilistic motion estimation based on temporal coherence

We develop a theory for the temporal integration of visual motion motivated by psychophysical experiments. The theory proposes that input data are tem porally grouped and used to predict and estimate the motion flows in the im age sequence. This temporal grouping can be considered a generalization of the data association techniques that engineers use to study motion sequence s. Our temporal grouping theory is expressed in terms of the Bayesian gener alization of standard Kalman filtering. To implement the theory, we derive a parallel network that shares some properties of cortical networks. Comput er simulations of this network demonstrate that our theory qualitatively ac counts for psychophysical experiments on motion occlusion and motion outlie rs. In deriving our theory, we assumed spatial factorizability of the proba bility distributions and made the approximation of updating the marginal di stributions of velocity at each point. This allowed us to perform local com putations and simplified our implementation. We argue that these approximat ions are suitable for the stimuli we are considering (for which spatial coh erence effects are negligible).