LoRa-Based Indoor Positioning in Dynamic Industrial Environments Using Deep Gaussian Process Regression and Temporal-Based Enhancements

Achieving precise localization in industrial settings presents significant challenges due to dynamic movements, complex layouts, and harsh environmental conditions that cause signal interference and reflections. This requires developing advanced indoor positioning systems that can handle these challenges and perform reliably even in the presence of dynamic movement. In this paper, a novel LoRa-based indoor positioning system designed for dynamic motion in industrial environments is presented. The proposed system integrates LoRa technology with a fingerprinting approach that involves fingerprint collection using the constant motion method and leverages a two-layer Deep Gaussian Process Regression (DGPR) model to overcome the non-linearity characteristics of signal propagation. Through testing on static and motion datasets, it was observed that collecting data in motion yields superior results for tracking dynamic objects. Furthermore, temporal-based enhancements like Temporal Weighted RSSI Averaging and Kalman filtering were introduced. These techniques effectively mitigate RSSI temporal variations and improve the reliability of position estimates. The experimental results, conducted in a real industrial environment, demonstrate that the proposed system achieves a mean positioning error of 1.94 meters and a 90th percentile error of 3.28 meters. These findings highlight the potential of combining LoRa technology with advanced machine learning algorithms and filtering techniques to achieve precise and reliable indoor tracking.