Can AI Replace Conventional Markerless Tracking? A Comparative Performance Study for Mobile Augmented Reality Based on Artificial Intelligence

Roberto Pierdicca, Flavio Tonetto, Marco Mameli, Riccardo Rosati, Primo Zingaretti

Abstract: AR is struggling to achieve its maturity for the mass market. Indeed, there are still many challenging issues that are waiting to be discovered and improved in AR related fields. Artificial Intelligence seems the more promising solution to overcome these limitations; indeed, they can be combined to obtain unique and immersive experiences. Thus, in this work, we focus on integrating DL models into the pipeline of AR development. This paper describes an experiment performed as comparative study, to evaluate if classification and/or object detection can be used an alternative way to track objects in AR. In other words, we implemented a mobile application that is capable of exploiting AI based model for classification and object detection and, at the same time, project the results in AR environment. Several off-the-shelf devices have been used, in order to make the comparison consistent, and to provide the community with useful insights over the opportunity to integrate AI models in AR environment and to what extent this can be convenient or not. Performance tests have been made in terms of both memory consumption and processing time, as well as for Android and iOS based applications.