Adaptive multi-feature motion estimation

A general discrete-time, adaptive, model-based motion estimation framework is introduced. The model is built to describe the dynamics of a set of points belonging to the same object over long periods of time. It incorporates two motion components, one for capturing the slowly-varying motion modes of the object and the second for accounting for the external forces and any other rapidly-changing motion modes, respectively. Time-varying mechanisms are also provided to handle occlusions or exits from the field of view. Applications of the motion model include temporal decimation/interpolation and noise filtering. In particular it is shown that the model provides a flexible framework for filtering out uncertainties in the feature matches.<>