A Novel Picture Fingerprinting Technique to Provide Practical Indoor Localization for Wheelchair Users

Abstract

Although GPS-based localization systems are widely used outdoors, they are inapplicable in indoor settings, where GPS signals are mostly unavailable. Existing research, however, has not found a feasible solution to obtain indoor positioning information. Existing indoor localization approaches either suffer from low accuracy or rely on expensive infrastructure and dedicated devices. In this study, we propose a novel picture fingerprinting-based indoor localization approach for power wheelchair users, who suffer from more restricted mobility. A location is represented by pictures (i.e., picture fingerprints) taken from different angles at the location. Localization is achieved by matching a locating picture, taken during wheelchair navigation, with the picture fingerprints of locations in a building. State-of-the-art deep learning techniques were employed to match locating and fingerprinting pictures. Experimental results showed that the proposed approach achieved accurate location recognition. Compared to existing indoor localization approaches, our proposed approach neither relies on dedicated infrastructures and devices nor requires labor-intensive maintenance and, therefore, provides a new direction for achieving feasible and accurate indoor localization.