Deep Learning-Based Fusion of Visible Light Positioning and IMU Sensors

Indoor Positioning System (IPS) allows us to precisely control robots and cars, and to improve the location awareness for pedestrians in indoor area where Global Positioning System fails. Many indoor positioning techniques leverage wireless signals, such as WiFi and Bluetooth, but suffer from low accuracy. Therefore, fusion techniques have been proposed to improve the accuracy, such as using the inertial measurement unit (IMU). In this paper, we proposed a deep learning-based fusion system using IMU and Visible Light Positioning (VLP). Different from WiFi and Bluetooth systems, VLP can provide highly accurate and fast location service with several line-of-sight (LOS) visible lights due to the special characteristic of light. Although VLP could fail in non-line-of-sight (NLOS) condition, the fusion system that we proposed is based on several structured neural networks and can carefully select VLP signals to achieve better positioning results even under poor LOS conditions. Based on our simulation results, the system can achieve an average of 20 cm accuracy.