Ahmad Keshavarz

Localization in LoRaWAN networks using the Received Signal Strength Indicator (RSSI) is widely adopted due to its low cost and extensive coverage. However, its accuracy is often compromised by environmental factors, such as entropy introduced by physical entities, which cause unpredictable fluctuations in RSSI values. Fixed fingerprint models struggle to adapt to these dynamic changes, further degrading localization performance. This paper introduces a novel approach leveraging CCTV images to enhance localization accuracy by dynamically updating the fingerprint map in real time. The proposed method utilizes environmental images to detect changes and identify prominent displaceable entities (PDEs), enabling adaptive updates to the pre-registered fingerprint map. The approach follows a two-step process: first, spectral graph clustering is employed to detect environmental changes by analyzing the relative positions of PDEs, gateways, and end nodes. Second, the impact of object distance on RSSI variations is evaluated through ANOVA analysis to refine the fingerprint map. The method was evaluated using data from three LoRa gateways in a real outdoor testbed. Experimental results demonstrate that the proposed image-based approach reduces localization error by an average of 53.2%, achieving a range of 32–65 meters across different days. In comparison, traditional methods such as Support Vector Machine (SVM) and Path Loss (PL) models exhibit errors ranging from 90–117 meters and 70–215 meters, respectively. These findings highlight the effectiveness of integrating image analysis into RSSI-based localization in LoRaWAN networks.