Author(s): P Pavani Sri Katyayini, Meghana Y, P Kusuma Sri, M Yedukondalu and M Gayathri

Abstract: Indoor localization is a critical application in smart spaces where GPS is simply not reliable. Wi-Fi fingerprinting is the most popular technique for this purpose, as it utilizes the already present wireless infrastructure and does not require any additional hardware. However, traditional fingerprinting techniques usually fail in the presence of dynamic indoor environments and random signal patterns. To address these challenges, a scalable localization framework combining machine learning and deep learning is proposed. At its foundation is a deep similarity network that translates RSSI fingerprints into an embedding space where similar geometric locations are close by. After the fingerprints are embedded, a machine learning inference task is used to compare the current fingerprint with the precomputed reference embeddings to deduce the most probable location. However, localization is not the only problem that needs to be solved. The system also contains a real-time signal monitoring component. It observes RSSI variations and connectivity status during active Wi-Fi scanning. A real-time dashboard is then used to display system status, signal behavior, and localization updates as they occur. Experimental evaluations demonstrate that the system provides reliable localization predictions and scales the fingerprint database over time without requiring any additional training of the model.

DOI: https://www.doi.org/10.33545/26633582.2026.v8.i2a.271

Pages: 10-15 | Views: 194 | Downloads: 49

Download Full Article: Click Here