Smartphone Indoor Positioning through Coordinate Alignment and Error State Fusion of Visual Inertial and WiFi Estimates

Abstract ntegrating visual-inertial and wireless position estimates requires a common coordinate frame and a fusion model that accommodates different update rates. This paper presents a position-level fusion method for smartphone indoor positioning that combines a visual-inertial navigation system (VINS) with Wi-Fi fingerprinting. A least-squares rigid transformation aligns the local VINS trajectory with the Wi-Fi fingerprint coordinate frame, and an error-state extended Kalman filter (ES-EKF) integrates the aligned VINS estimates with intermittent Wi-Fi position observations. Data were collected using a Xiaomi Mi 9 smartphone and processed offline with VINS-Mono. Five repeated trials were conducted in each of two scenarios with trajectories of approximately 50 m and 150 m. In the short-trajectory experiment, VINS+Wi-Fi achieved a root mean square error (RMSE) of 0.3429 ± 0.038 m, compared with 0.3821 ± 0.052 m for VINS and 0.7468 ± 0.088 m for Wi-Fi, a 10.3% reduction relative to VINS. In the longtrajectory experiment, the corresponding RMSE values were 0.5909 ± 0.079 m, 0.8006 ± 0.121 m, and 2.1626 ± 0.345 m, a 26.2% reduction relative to VINS. Values are reported as means and standard deviations across trials. These results support the use of coordinate alignment and asynchronous position-level fusion to reduce positioning error under the tested static, single-floor conditions. The prototype required eight Wi-Fi access points and a preconstructed fingerprint database; real-time on-device processing was not evaluated.