Vision-Based Indoor Corridor Localization via Smartphone Using Relative Distance Perception and Deviation Compensation

Nowadays, the determination of the indoor position has become a hot issue in industry and academia as it is fundamental for many applications and services. Thanks to the pervasive availability of smartphones, a plethora of indoor navigation systems has been developed based on different techniques, such as radio-frequency-based technologies, sensor-based technologies, and vision-based technologies; however, there are certain limitations for general use, such as high cost of adding extra hardware, more labor input, violent fluctuation, low accuracy result, etc. To improve the robustness and accuracy of traditional means, this paper proposes a new solution to accurate indoor positioning for narrow corridors. The proposed method consists of coarse positioning from Wi-Fi matching and image-level positioning and verification from holistic and local visual feature matching. A rough position area called subarea is first predicted using Wi-Fi information and a geometric relationship derived in this paper is used to calculate user’s coordinates in higher precision. A feed-forward neural network is also trained for error mitigation. The experiments conducted in this paper have proved that our method owns an above-average precision (positioning error results fewer than 0.3 m on average) in related field which can be used to locate users in corridors or channels accurately while still keeping humble in labors.