Reducing Information Overload with Machine Learning in Mobile Pervasive Augmented Reality Systems

Augmented reality systems in dynamic environments still struggle with the challenge of what information should be displayed at which time. This work focuses on the case of Mobile Pervasive Augmented Reality Systems (MPARS) and their use in dynamic environments such as outdoor sports. An open-source proof-of-concept for a machine learning-based architecture to implement an MPARS on a specific use case of outdoor usage in a sports environment is presented. The new design for the system relies on heuristics that combine technology acceptance indicators, sensing, and information volume criteria to show the user a contextually meaningful subset of information. The information to the user is displayed in close-to-real-time, and the system can adjust and customise to prevent information overload. A first set of experiments was carried out based on end-user preferences to show the feasibility of the proposed system. To provide meaningful feedback, i.e., the right information when needed or wanted, to sports users on their MPARS experience, a predictive model was trained and shown to be able to estimate when information should be displayed to the user.