Study On Key Technology Of Wide Area Localization And Registration For Mobile Augmented Reality Systems

With the wide use of smart phones, researchers have drawn more and more attention to Mobile AR. Compared with the PCs, the mobile devices such as smart phones have limited resources such as low computation power, limit memory size, low power consumption and so on. Meanwhile, smart phones are small and portable, which widens people’s moving area, so the on-devices wide area localization and registration for mobile AR is required. Since the technology of AR on PCs cannot be directly transplanted on the smart phones, this paper proposes a system of on-devices wide area localization and registration for mobile AR.The main achievements of this paper are as follows:First, we use gravity information to enhance the discriminative power of the original VLAD descriptors and design an effective vector quantization strategy, which can compress high-dimensional visual descriptor into compact codes with several bytes to facility the searching process. We also design an image searching engine which fits the mobile devices. We also construct a vision and senor based compressed index structure which can directly deal with wide area image database.Secondly, we design a simple and effective vector binaryzation method to reduce the usage of memory and propose a location aware algorithm to fuse multiple features. This algorithm can compress multiple features into a compact high discriminative descriptor. We also design a method to optimize feature fusion and index structure jointly in order to obtain accurate and memory efficient location recognition.Thirdly, we design a fast camera initializing and tracking method. The result is a system that can track non planar scenes at frame rates of up to10Hz on current generation phones, which solves the problem of the registration of wide area mobile AR.Fourth, we also release a new database which contains1.295million geo-tagged street view images and849testing queries. We think that the database can be used as a new benchmark for future research on related topics.The experiments of this paper prove that the on-devices wide area location and registration of mobile AR system we propose can improve the recognition and registration accuracy, save the memory and reduce response time, which provides the key technology to realize mobile AR in daily use.