Please use this identifier to cite or link to this item:
Title: Tracking for Mobile 3D Augmented Reality Applications
Keywords: Augmented Reality, computer vision, inertial, GPS, hybrid tracking
Issue Date: 29-Sep-2009
Source: FONG WEE TECK (2009-09-29). Tracking for Mobile 3D Augmented Reality Applications. ScholarBank@NUS Repository.
Abstract: This thesis presents the PhD research carried out on tracking for mobile 3D augmented reality applications. Augmented Reality (AR) is the superposition of the virtual and the real environments, so that both the virtual elements and the real world can be interactively perceived by the user. The research focus is on robust, wide-area tracking for high precision 3D AR applications in non-prepared environments. The main motivation is to move AR out of laboratory, so as to achieve mobile AR in outdoor environments. Tracking allows the AR system to determine the segments of the real world that the user is looking at, so that virtual 3D objects can be inserted to appear visually coherent to the user. This allows computer systems to augment the mobile user¿s reality. Wide-area applications require the tracking systems to operate in a wide range of conditions, and over a wide range of motion. Robustness, precision, low latency and jitter are important requirements for successful and satisfactory augmentation. This research takes a multidisciplinary approach towards solving the research question, through investigating three different but complementary tracking systems, namely Computer Vision (CV), Inertial Measurement and Global Positioning Systems (GPS), to derive a hybrid wide-area tracker, known as the Augmented Reality TRackIng SysTem, or ARTIST. This approach is chosen based on the observation that no single property and sensors is able to meet the requirements of robustness and precision. Sensors with complementing strengths and weaknesses can be combined together to approximate a perfect sensor. As Inertial and GPS function well over a large area, the approach taken is to first improve the precision of both sensors, so that they can work reasonably well in regions where CV fails. The research on inertial measurement focused on the calibration of MEMS-based sensors, so that they can be used as independent orientation trackers. Calibration methods for tri-axial accelerometers and gyroscopic systems that are completely independently of external equipment have been developed. This allows end-users to perform calibration on-site, which has not been achieved for gyroscope calibration. For GPS, a novel method for GPS positioning based on the Differential Single Difference of GPS carrier phase measurement has been formulated. It is suitable for AR positioning with an accuracy of 10 cm, and avoids the computationally expensive resolution of integer ambiguity. However, the level of precision achieved is not comparable to CV. The research on CV focuses on marker-less 6DOF tracking using natural features. A CV tracker with accurate 3D augmentation and good robustness against illumination changes, partial occlusion and extreme object poses, has been developed and tested. Simultaneous augmentation of three objects at 15fps was achieved through efficient system design, as well as improvements to underlying algorithms. Specifically, the keypoint signature is improved with a proposed matching method, which maintains the matching accuracy with lower computation load. New models were also proposed for Efficient Second-order Minimization (ESM) that allows for handling of radial distortion, shadows, specular glares and partial occlusion. Finally, two methods are developed to combine the sensor output. The first is a loosely coupled configuration where standalone GPS and inertial measurements are used to limit the search set for initializing the computer vision tracker. This system only works in areas where there are sufficient features for CV tracker. The second is a Kalman Filter based hybrid tracker, where low level sensor outputs, consisting of differential GPS carrier phases, acceleration and angular velocity and CV measure are combined. The second system is a true wide area tracker, with degraded precision with GPS and inertial tracking when CV fails.
Appears in Collections:Ph.D Theses (Open)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
FongWT.pdf4.31 MBAdobe PDF



Page view(s)

checked on Dec 11, 2017


checked on Dec 11, 2017

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.