Improving Indoor Navigation Accuracy with Neural Networks: A Focus on Signal Propagation Challenges

The purpose of this project is to increase the accuracy and robustness of Indoor Positioning Systems (IPS) in a four-story office and school building by utilizing neural networks. The complex signal propagation in traditional IPS methods, which rely on RSSI values collected from Wi-Fi and Bluetooth Beacons, presents challenges. In contrast to prior methodologies, the neural network model demonstrated enhanced efficacy when trained with modern techniques such as dropout and L2 regularization. The system's average location error was 1.3 meters, while trilateration and fingerprinting had errors of 3.7 meters and 2.6 meters, respectively. The model demonstrated strong adaptability to changes in the environment and had little latency, making it highly suitable for real-time applications. This research highlights the ability of neural networks to provide more precise and reliable solutions for indoor positioning.