Real-time camera tracking in previously unknown scene is attractive to a wide spectrum of computer vision applications. In Recent years, Simultaneous Localization and Mapping (SLAM) system and its varieties have shown extraordinary camera tracking performance. However, the robustness of these systems to rapid and erratic camera motion is still limited because of the typically used Local Localization scheme. To overcome this limitation, we present an efficient online camera tracking algorithm using a Global Localization scheme which matches features in a global way through two steps: First, coarse matches are obtained through nearest feature descriptor search. Afterwards, a Game Theoretic approach is exploited to eliminate the incorrect matches and the left correct matches can be used to estimate the camera pose. Result shows our camera tracking algorithm has significantly improved the robustness of camera tracking system to rapid and erratic camera motion.
展开▼
