Evaluating and enhancing google tango localization in indoor environments using fiducial markers

• Published in: 2018 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC) pp. 142 - 147
• Main Authors: Marques, Bernardo, Carvalho, Raphael, Dias, Paulo, Oliveira, Miguel, Ferreira, Carlos, Sousa Santos, Beatriz
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.04.2018
• Subjects: 3D indoor navigation; Area learning; Augmented Reality; Lighting; Markers; Mobile handsets; Motion tracking; Sensors; Switches; Tango platform; Three-dimensional displays; Tracking; Visualization
• DOI: 10.1109/ICARSC.2018.8374174

Recent advances in 3D sensing technologies, as well as in inertial measurement technologies, have resulted in significant improvements in the accuracy of the localization of systems that combine all these sensors. Project Tango is one of the most successful examples of such systems. Developed by Google, it integrates in an Android mobile device a set of sensors and software required to provide accurate real-time 3D information when moving the equipment freely in hand. This is making mapping and navigation accessible to the general public, with evident applications in robotics, augmented reality, computer vision and others. The contribution of this paper is towfold: first, we present a thorough evaluation of the localization accuracy of the Tango platform in different conditions; second, we present a fiducial marker-based extension of the Tango localization system, which improves the localization estimates in certain conditions. The paper presents a set of experiments performed to evaluate the position and orientation errors in indoor environments, using Augmented Reality for visualization purposes, with and without area learning, e.g. using a priori information acquired from the environment. In addition, we propose a solution based on the use of additional visual markers, which allows the re-calibration of augmented content in specific locations, to improve tracking accuracy in dynamic environments where spatial and/or illumination changes may occur. A statistical analysis of the results shows that the Tango with area learning and the proposed solution provide a level of accuracy significantly better that the Tango without area learning. Moreover, the proposed solution can overcome some limitations of Tango with area learning when used in dynamic environments.