Wide area localization and tracking on camera phones for mobile augmented reality systems

摘要

We design a computer vision-based localization and tracking method to cope with the wide area registration problem in mobile AR applications. To get an efficient wide area localization system, we design a fast and compact key frame search algorithm using the modified vector of locally aggregated descriptors and the improved transform coding-based approximate nearest neighbor search approach. While providing reasonable search accuracy, the proposed algorithm can also reduce the memory usage on mobile phones significantly. To get a real-time registration system, we use a flexible camera initialization and tracking method. The method can track non-planar scenes at frame rates of up to 10 Hz on current generation phones and facilitate the registration process of wide area mobile AR applications to a large degree. Some results are discussed to prove the effectiveness of the proposed localization and tracking method for wide area mobile AR applications.