Indoor localization with a signal tree

Smartphones embedded with cameras and other sensors offer possibilities to attack the problem of indoor localization where GPS is not reliable. In this paper, a novel tree-based localization system is proposed based on WiFi, inertial and visual signals. There are three levels in the tree: (1) WiFi-based coarse positioning. The WiFi database of a building is clustered into several branches for coarse positioning; (2) Orientation pruning. Images collected in a building are tagged with camera orientations towards which they are taken, so when inferring a user’s location by comparing the query image the user takes with the reference image dataset, the image branches tagged with unmatched orientation will not be searched; (3) Fine visual localization. The user’s location is accurately determined by matching the query image with the reference image dataset based on a multi-level image description method. Our signal tree based method is compared with other methods in terms of the localization accuracy, localization efficiency and time cost to build the reference database. Experimental results on four large university buildings show that our indoor localization system is efficient and accurate for indoor environments.