Photo-shoot localization for mobile AR based on registered images in virtualized reality models
In Workshop KJMR 2013, The 6th Korea-Japan Workshop on Mixed Reality. 2013.
Jun Nishida, Koji Makita, Tomoya Ishikawa, Takashi Okuma, Masakatsu Kourogi, Thomas Vincent, Laurence Nigay, Jun Yamashita, Hideaki Kuzuoka, Takeshi Kurata
Résumé
This paper presents a study on estimating photo-shoot location and orientation in indoor environments for augmented reality applications. The proposed method is based on registered images in virtualized reality models [1]. Registered images are secondary products of model creation, and contain photo-shoot location, orientation, and depth information. Therefore, additional data for special purposes need not be created. The upper and lower left parts of Figure 1 show the system architecture and data flow. We assume that the proposed method is used in combination with a relative positioning system, for example, Pedestrian Dead Reckoning (PDR). In the proposed system, a real image taken by a mobile device is sent to a server, and the image is compared with the registered images.
In this study, we compare the localization performance of the proposed system using mutual information, edge information, SURF, ORB, and FREAK features.