Scalable recognition and tracking for mobile augmented reality

Abstract

For an augmented reality application to be realistic, exact tracking of target objects is essential. However, recent mobile augmented reality applications such as location or recognition-based applications lack realism in augmentation due to inexact tracking methods. Visual tracking based tracking is capable of being exact and robust, but in a mobile augmented reality system, the number of objects that can be augmented is far limited. In this paper, we propose a new framework that overcomes the limitations of the previous works. First, our framework is scalable to the number of objects being augmented. Second, our framework provides an improved and realistic augmentation by adopting a real-time accurate visual tracking method. To the best of our knowledge, there has been no system proposed successfully integrating both of these properties. To achieve scalability, the bag of visual words based recognition module with a large database runs on a remote server and the mobile phone tracks and augments the target object by itself. The server and mobile phone is connected by conventional Wi-Fi. Including network latency, our implementation takes 0.2 seconds for initiating an AR service on a 10,000 object database, which is fairly acceptable for a real-world augmented reality application.