Augmented Reality with Android

Student: Pedro Carvalho
Supervisor: António Ramires

Abstract

Augmented Reality (AR) relies on overlaying 3D objects over a real environment on a screen display. Recently this technology has begun to gain a lot of attention by the common consumer because it is starting to be used in devices available to them, one of those devices being mobile smartphones. Hence, it is relevant to seek the limits and explore the potential that these devices are capable of, in order to build a balanced AR experience.

The purpose of this work was to study the limits of developing a localization technique for an Android smartphone which was accurate and fast. This technique was to be used in an AR application which allowed the user to view large 3D environments related to the real scene (e.g a room could be turned into an aquarium).

The research was initially targeted at finding the right technique to localize the device, by assessing the accuracy of algorithms like Inertial Navigation Systems, Monocular Visual Odometry and SLAM on a hardware limited device – a smartphone. The results obtained with these algorithms did not meet the precision requirements to avoid misalignments between virtual and real entities.

The solution found was to use 2D markers around the real scene with the help of a computer tool which comprised a novel feature which aided the user in positioning the markers with visual cues painted on the 3D model’s surface.

Markers detected by the smartphone’s camera allowed the computation of the position of the device. These markers are detected using Vuforia’s SDK Image Target Recognition. Furthermore, some improvements were implemented to provide more robust tracking for large 3D Models.

A framework was devised which consists of the android AR application and the computer tool which, aside from helping the user position the markers, can build AR setups and deploy them to the mobile app.

Finally, some tests were made which assess the accuracy of the implementation. Example AR scenes were built to examine how the application fares in very different environments, one being a room and the other a public square and streets in a village.

Thesis Download (PDF)