Tracking points using projective reconstruction for augmented reality

Abstract

Natural feature tracking is a popular research topic in computer vision and has been used in Augmented Reality (AR) applications. Current natural feature trackers have two major limitations in AR applications. Firstly, the natural features may be lost during the tracking process such that the annotations that are linked to these natural features will also be lost. Secondly, if the virtual objects have to be augmented on regions where there are no distinct natural features that can be tracked, the current natural feature cannot be used directly. This paper proposes a method for points tracking or transferring based on the Kanade-Lucas-Tomasi (KLT) feature tracker and the projective reconstruction technique. The points to be tracked include the lost natural features, and any points that are specified by the users. The proposed method is useful for AR applications, including scene annotation, registration, etc. Some indoor and outdoor experiments have been conducted to validate the performance of the proposed method. Copyright © 2005 by the Association for Computing Machinery, Inc.