Motion reconstruction using moments analysis

This work presents a method for motion reconstruction. The approach is based on image moments and the bsp-tree data structure. We use invariant properties of these moments to construct a bsp-tree and determine a set of ellipses that approximates the object's shape in each frame. These ellipsoidal structures are matched, which allow us to track the object. Motion is represented by the geometric transformations between the sets of ellipses and stored in the hierarchical structure. Motion reconstruction is done by interpolating these transformations.