Unsupervised Manifold Alignment for Wifi RSSI Indoor Localization

Abstract

There is a wealth of analysis techniques that researchers across the world are implementing for better indoor localization. The RSSI fingerprinting is one of many techniques used for indoor and outdoor localization. In addition, other fingerprints are used to assist in the localization collected from several sources such as camera, radar, and Lidar. Ideally, a combination of these sources is used to locate the same object. Precisely, these sources are collecting the same data using different dimensions ultimately building upon one big system. Due to different dimensions set by these sources, it often becomes difficult to train the overall system to achieve the task of localization. In this paper, we propose a technique that can be used to incorporate training multiple datasets from different dimensions (e.g. Lidar, camera, and radar) into one global dataset, then train it all at once. This technique is known as the Manifold Alignment. Our proposed manifold alignment algorithm bridges the gap, allowing the inclusion of multiple datasets in our application whilst constraining the computational time and storage that would be required for the system. We assume that our technique is embedded into our system and localization is achieved through either computing our proposed Manifold Alignment algorithm over a local device, edge server, or cloud. Results in this paper show how well the Manifold Alignment Algorithm is beneficial for a localization problem where it is implemented inside a machine learning model that computes the manifold of these datasets.