Wide area localization on mobile phones

摘要: We present a fast and memory efficient method for localizing mobile user's 6DOF pose from single camera image. Our approach registers view with respect to sparse 3D point reconstruction. The dataset is partitioned into pieces based on visibility constraints occlusion culling, making it scalable handle. Starting coarse guess, our system only considers features that can be seen the position. resource efficient, usually requiring few megabytes of memory, thereby feasible run low-end devices such as phones. At same time enough give instant results this device class.