Accurate On-Body Distance Estimation using BLE RSSI and IMU Sensor Fusion

Abstract

Signal strength-based localization has gained significant traction, with technologies like Wi-Fi, BLE, and UWB being widely studied for indoor applications. However, the specific use of BLE Received Signal Strength Indicator (RSSI) for robust, short-range distance estimation on the human body remains a challenging area, primarily due to signal instability caused by multipath fading and body shadowing. This thesis presents a novel framework that fuses BLE RSSI with Inertial Measurement Unit (IMU) orientation data to overcome these limitations. We developed a custom wearable system using ESP32 microcontrollers and BNO085 IMUs and conducted a study with 6 human participants performing a series of defined arm movements. A high-precision XSens motion capture suit provided ground-truth data. This data was used to train a Long Short-Term Memory (LSTM) based model, termed QuaternionRNN, which leverages temporal sequences of normalized quaternions and RSSI to predict inter-sensor distances. The proposed model achieved a median absolute distance prediction error of approximately 3 cm on a generalized dataset. Furthermore, in a rigorous subject-wise cross-validation analysis, the model demonstrated strong generalization to unseen users, maintaining a median absolute error between 4-7 cm. This work establishes that fusing BLE RSSI with IMU data provides a cost-effective, accurate, and generalizable solution for on-body localization, offering a viable tool for applications in remote rehabilitation monitoring and human motion analysis.