Browsing by Subject "environment"

Enabled by decades of pioneering research, in the last few years markerless mobile augmented reality (AR) has become widely available on commercial devices, and we are tantalizingly close to realizing the benefits of this technology in numerous industries. However, a lack of robustness across diverse real-world environments remains a stubborn barrier to greater adoption. Specifically, while the techniques underlying spatial registration of virtual content perform well in ideal conditions, commonly encountered environments result in virtual content spatial artifacts, which severely degrade a user's sense of immersion and their ability to interact with virtual content. The fundamental nature of this problem, combined with the restrictions imposed by mobile devices, means that this issue is unlikely to be resolved in the near future. Therefore, we require solutions that guide AR users and designers towards environments or virtual content placements that minimize spatial artifacts, without compromising other factors which contribute to a user's quality of experience.

To this end, this dissertation considers challenges associated with developing environment analysis and design systems for markerless mobile AR. Our contributions address the challenges of quantifying virtual content stability artifacts, estimating and visualizing pose errors that cause those stability artifacts, designing environments that take into account multiple system and human factors, and creating optimal environment textures.

We start with our work on quantifying virtual content stability artifacts on commercial AR platforms. To enable the collection of data in a variety of real environments, we developed a new virtual content stability measurement method, which we incorporated into a cross-platform, open-source AR session measurement app. We used this app to perform the first direct, quantitative comparison of virtual object stability on both multiple smartphone models and an AR headset. Our results illustrated the widespread nature of stability artifacts, especially on smartphones, and provided insights on the influence of device hardware on the magnitude of artifacts. We also conducted studies on user perception of virtual content spatial artifacts, which indicated that they are a frequent and noticeable issue.

We then describe our efforts on estimating the pose tracking errors that cause stability artifacts, and visualizing these estimates in a way that is actionable for AR users and environment designers. We developed the first uncertainty-based pose error estimation method for VI-SLAM, which classified pose error magnitude with an accuracy of 96.1%. We implemented this method for real devices in the first situated trajectory analysis system for AR that incorporates pose error estimates, and publicly released the code required to implement it. We designed and evaluated three pose error visualization techniques, which revealed the impact of the environment on their efficacy. We also developed proof-of-concept systems for machine learning-based environment classification and online pose error estimation.

Next, we present our integrated design methodology for environments which host AR, which takes into account both system and human factors. We enabled this by developing a technique for the use of virtual environments in pose tracking evaluations, with the rationale that they facilitate full control over conditions, as well as virtual reality (VR)-based evaluations of user satisfaction. We used this technique for systematic experiments on the effect of environment texture, revealing that complexity and edge strength are desirable properties for pose tracking. We showcased our overall methodology with a case study on AR museum design. We further evidenced the efficacy of virtual environments by creating a VI-SLAM point cloud object detection dataset, and demonstrated the application of integrated design principles in a proof-of-concept environment illuminance optimization system.

Finally, we detail our in-depth examination of how environment textures can be optimized for AR experiences. We conducted the most extensive evaluation of the effect of environment texture on VI-SLAM performance to date, with over 5000 total trials. Our results demonstrated the potential for low-contrast invisible textures, which result in good pose tracking performance without impairing human perception of virtual content. We conducted a study that showed an invisible environment texture resulted in lower user workload than a complex texture in an AR assembly task. We then evaluated invisible textures in more mobile scenarios, which revealed the challenges associated with certain types of device motion. We used these findings to design the first multiuser motion mapping system for mobile AR, and evaluated a texture produced using its guidance in a realistic case study.