Image Recognition-Based Indoor Positioning System Using Federated Learning

With the advancement of artificial intelligence (AI), indoor positioning systems have become increasingly important for various applications, leading to the development of diverse indoor positioning methods. Among various approaches, image-based indoor positioning methods have demonstrated relatively good positioning accuracy. However, maintaining user privacy continues to be a significant challenge. To tackle this issue, we propose a Federated Learning (FL) approach and develop an FL-based indoor image recognition positioning system. We conducted experiments to validate our method using two real-world datasets and compare the results to a Non-FL approach. Furthermore, we evaluated the accuracy of our method across various communication rounds, the number of client devices, and the amount of data per client device. Our experimental results indicate that the proposed method effectively preserves client privacy while achieving accuracy similar to the Non-FL approach. By employing FedAvg and FedOpt algorithms with the MobileNet model, our system ultimately reaches 94% accuracy.