Performance Enhancement of the Vision System for the Augmented Reality Thymio Robot

Abstract

The aim of this thesis is the improvement of the vision system of the Thymio Programming Adventure, an application which uses augmented reality to help children discover programming by letting them control an educational robot through a virtual world. The improvement is about the handling of multiple markers observed by the application and the use of its position to augment the observed world with virtual objects. In particular, the implementation of an appropriate data structure shall improve the vision system for cases where a strict subset of the markers is occluded. As a supplementary objective, experiments with the purpose to gather insights into the existing vision system are conducted.

To start with, a detailed statement of the problem is followed by a discussion of theoretical aspects important for this project. The thesis then presents the implementation of a data structure able to store and retrieve transformations followed by the description how this data structure can be integrated into the existing application to solve the stated problem. Lastly, the execution and results of experiments investigating the quality difference of the used markers and the functionalities of the implemented data structure can be found.

The implementation of the data structure presented does improve the vision system of the application. It allows to describe all virtual objects relative to one marker. Its position is known as long as at least one of multiple markers is visible, and therefore no reordering of the scene graph is necessary regardless what strict subset of markers is occluded. Through the conduction of experiments, a quality difference between the used markers is identified. The author recommends further investigation to determine the reason for the disparity since the quality of the marker is essential for the performance of the vision system.