Real-Time and Markerless Vision-Based Tracking for Outdoor Augmented Reality Applications

In this paper, a new concept for markerless optical tracking called “tracking with reference images” is introduced . This concept provides a flexible and praticable framework and is especially relevant for outdoor augmented-reality applications, for which no tracking solution currently exists. The implementation is achieved using an image matching technique, which compares the current live video image with one or more of the reference images. The complete system has been tested outdoor in the context of a mobile AR-application for archeology. The system runs on a laptop at around 10 Hz and provides views of virtual monuments superposed to their ruins. 1 Optical tracking with reference images Markerless optical tracking is a very complex task. One can apply an image processing operator, like the LucasTomasi or Harris corner detector, to the video images and recover the camera trajectory out of the motion of the 2D features. This approach works well off-line because all features are accessible at a time, what enables filtering or minimization of the errors by e.g. bundle-adjustment techniques. Beyond it, one important characteristic of this kind of approaches is, that the motion is given in an arbitrary coordinate system and with an arbitrary scale because all information are got only from the images. On contrary, realtime Augmented-Reality (AR) applications require absolute coordinates, usually provided by the markers of an optical tracking-system. We propose here to substitute the markers with another support, namely with standard images of the environment. These images can be considered as the (necessary) reference component of the tracker and are called for this reason “reference images”. 0.3 0.7