Indoor Positioning using Adaptive Fusion for WiFi and IMU under Android Scan Throttling

Abstract

Indoor localization remains a challenging problem in modern environments. Interference caused by building structures, ceilings, and floors makes it impossible to reliably localize by satellite radio-signals. Recent approaches to WiFi fingerprinting range from traditional distance-based methods to more complex machine learning models, such as Convolutional Neural Networks (CNN) and Transformers, to achieve robust results. However, they perform best at frequent WiFi scanning and degrade when readings are sparse. Another method is to take advantage of the Inertial Measurement Unit (IMU) sensors' high-frequency motion information for trajectory predictions. IMU-based localization, however, comes with the downside of drift accumulation. This thesis explores the capabilities of an IMU + WiFi fusion-based method of Indoor localization under WiFi scan throttling. Modern phones put restrictions on high-frequency WiFi scans. This thesis uses IMU-based PDR with corrections from a radiomap built on pre-collected WiFi data. Results demonstrate that IMU–WiFi fusion significantly improves localization accuracy under WiFi scan throttling, maintaining stable performance even with sparse WiFi updates.