City-scale location services based on mobile augmented reality

Mobile Augmented Reality system presents surroundings in a way of virtual-real fusion for location-based services. This paper employs C/S architecture for vision-based mobile AR system, which solves the problem of processing speed and storage space on mobile phone. Vocabulary tree algorithm which is based on distributed clustering algorithm is used for city-scale scene recognition on the server. BRISK feature and optical flow algorithm is used in combination for visual tracking on mobile devices. Experimental results on UKbench and SUN397 show that the proposed distributed clustering algorithm can quickly build vocabulary tree and solve the limitations of memory. Mobile terminals keep real-time tracking through extraction of BRISK feature and design of parallel pipeline.