Extendible object-centric tracking for augmented reality

Presents a novel object-centric tracking architecture for presenting augmented reality media in spatial relationships to objects, regardless of the objects' positions or motions in the world. The advance this system provides over previous object-centric tracking approaches is the ability to sense and integrate new features into its tracking database, thereby extending the tracking region automatically. This lazy evaluation of the structure-from-motion problem uses images obtained from a single calibrated moving camera and applies recursive filtering to identify and estimate the 3D positions of new features. We evaluate the performance of two filters; a classic extended Kalman filter (EKF) and a filter based on a recursive average of covariances (RAC). Implementation issues and results are discussed in conclusion.