Real-time 3D camera tracking for industrial augmented reality applications

In this paper we present a new solution for real-time 3D camera pose estimation for Augmented Reality (AR) applications. The tracking system does not require special engineering of the environment, such as placing markers or beacons. The required input data are a CAD model of the target object to be tracked, and a calibrated reference image of it. We consider the whole process of camera tracking, and developed both an autonomous initialization and a real-time tracking procedure. The system is robust to abrupt camera motions, strong changes of the lighting conditions and partial occlusions. To avoid typical jitter and drift problems the tracker performs feature matching not only in an iterative manner, but also against stable reference features, which are dynami-cally cached in case of high confidence. We present experimental results generated with help of synthetic ground truth, real off-line and on-line image sequences using different types of target objects.