Scalable Wi-Fi RSS-Based Indoor Localization via Automatic Vision-Assisted Calibration

Wi-Fi-based positioning promises a scalable and privacy-preserving solution for location-based services in indoor environments such as malls, airports, and campuses. RSS-based methods are widely deployable as RSS data is available on all Wi-Fi-capable devices, but RSS is highly sensitive to multipath, channel variations, and receiver characteristics. While supervised learning methods offer improved robustness, they require large amounts of labeled data, which is often costly to obtain. We introduce a lightweight framework that solves this by automating high-resolution synchronized RSS-location data collection using a short, camera-assisted calibration phase. An overhead camera is calibrated only once with ArUco markers and then tracks a device collecting RSS data from broadcast packets of nearby access points across Wi-Fi channels. The resulting (x, y, RSS) dataset is used to automatically train mobile-deployable localization algorithms, avoiding the privacy concerns of continuous video monitoring. We quantify the accuracy limits of such vision-assisted RSS data collection under key factors such as tracking precision and label synchronization. Using the collected experimental data, we benchmark traditional and supervised learning approaches under varying signal conditions and device types, demonstrating improved accuracy and generalization, validating the utility of the proposed framework for practical use. All code, tools, and datasets are released as open source.