2017 Theses Doctoral
Augmented Reality Interfaces for Enabling Fast and Accurate Task Localization
Changing viewpoints is a common technique to gain additional visual information about the spatial relations among the objects contained within an environment. In many cases, all of the necessary visual information is not available from a single vantage point, due to factors such as occlusion, level of detail, and limited field of view. In certain instances, strategic viewpoints may need to be visited multiple times (e.g., after each step of an iterative process), which makes being able to transition between viewpoints precisely and with minimum effort advantageous for improved task performance (e.g., faster completion time, fewer errors, less dependence on memory).
Many augmented reality (AR) applications are designed to make tasks easier to perform by supplementing a user's first-person view with virtual instructions. For those tasks that benefit from being seen from more than a single viewpoint, AR users typically have to physically relocalize (i.e., move a see-through display and typically themselves since those displays are often head-worn or hand-held) to those additional viewpoints. However, this physical motion may be costly or difficult, due to increased distances or obstacles in the environment.
We have developed a set of interaction techniques that enable fast and accurate task localization in AR. Our first technique, SnapAR, allows users to take snapshots of augmented scenes that can be virtually revisited at later times. The system stores still images of scenes along with camera poses, so that augmentations remain dynamic and interactive. Our prototype implementation features a set of interaction techniques specifically designed to enable quick viewpoint switching. A formal evaluation of the capability to manipulate virtual objects within snapshot mode showed significant savings in time spent and gain in accuracy when compared to physically traveling between viewpoints.
For cases when a user has to physically travel to a strategic viewpoint (e.g., to perform maintenance and repair on a large physical piece of equipment), we present ParaFrustum, a geometric construct that represents this set of strategic viewpoints and viewing directions and establishes constraints on a range of acceptable locations for the user's eyes and a range of acceptable angles in which the user's head can be oriented. Providing tolerance in the allowable viewing positions and directions avoids burdening the user with the need to assume a tightly constrained 6DOF pose when it is not required by the task. We describe two visualization techniques, ParaFrustum-InSitu and ParaFrustum-HUD, that guide a user to assume one of the poses defined by a ParaFrustum. A formal user study corroborated that speed improvements increase with larger tolerances and reveals interesting differences in participant trajectories based on the visualization technique.
When the object to be operated on is smaller and can be handheld, instead of being large and stationary, it can be manually rotated instead of the user moving to a strategic viewpoint. Examples of such situations include tasks in which one object must be oriented relative to a second prior to assembly and tasks in which objects must be held in specific ways to inspect them. Researchers have investigated guidance mechanisms for some 6DOF tasks, using wide--field-of-view (FOV), stereoscopic virtual and augmented reality head-worn displays (HWDs). However, there has been relatively little work directed toward smaller FOV lightweight monoscopic HWDs, such as Google Glass, which may remain more comfortable and less intrusive than stereoscopic HWDs in the near future. In our Orientation Assistance work, we have designed and implemented a novel visualization approach and three additional visualizations representing different paradigms for guiding unconstrained manual 3DOF rotation, targeting these monoscopic HWDs. This chapter includes our exploration of these paradigms and the results of a user study evaluating the relative performance of the visualizations and showing the advantages of our new approach.
In summary, we investigated ways of enabling an AR user to obtain visual information from multiple viewpoints, both physically and virtually. In the virtual case, we showed how one can change viewpoints precisely and with less effort. In the physical case, we explored how we can interactively guide users to obtain strategic viewpoints, either by moving their heads or re-orienting handheld objects. In both cases, we showed that our techniques help users accomplish certain types of tasks more quickly and with fewer errors, compared to when they have to change viewpoints following alternative, previously suggested methods.
- Sukan_columbia_0054D_13990.pdf application/pdf 11.9 MB Download File
More About This Work
- Academic Units
- Computer Science
- Thesis Advisors
- Feiner, Steven K.
- Ph.D., Columbia University
- Published Here
- July 30, 2017