SensNet: An End-to-End Deep Learning-based BLE-IMU Fusion Positioning System for Industry 4.0

Existing wireless-inertial fusion positioning mechanisms predominantly rely on empirical propagation models of wireless signals and are often fused by filtering algorithms such as the Kalman filter or particle filter. Despite their wide usage, these empirical models, especially in relation to system and noise, often exhibit lower accuracy in practical positioning scenarios. Recently, deep learning has been widely studied to build a more robust localization system along with filter fusion, which still needs empirical noise coefficients. To mitigate these constraints, this paper puts forward a novel fusion positioning system based on an end-to-end network, namely, 'SensNet'. The proposed approach deviates from the conventional reliance on empirical models and provides an innovative solution for positioning. The validity of SensNet is underscored by a comprehensive evaluation using the fusion of Bluetooth Low Energy and Inertial Measurement Unit (BLE-IMU) positioning in an indoor scenario. The results evidence an impressive mean positioning error for the fusion network of merely 0.415m, thereby highlighting the potential of SensNet as a reliable solution for Industry 4.0 positioning systems.