Learning Shape and Motion from Image Sequences

The node‐decoupled extended Kalman filter (NDEKF) algorithm is used to deal with high‐dimensional signals: moving visual images. Many complexities arise in visual processing that are not present in one‐dimensional prediction problems: the scene may be cluttered with background objects, the object of interest may be occluded, and the system may have to deal with tracking differently shaped objects at different times. This chapter looks at the problem of tracking objects that vary in both shape and location. A neural network model makes use of short‐term continuity to track a range of different geometric shapes. Three experiments are presented to evaluate the model's abilities.