ISRG Schedule, Spring 2008

Research in Virtual Reality typically involves dealing with a range of different devices. And since each device has different input configuration and communication interfaces, setting up an experiment using many of them may be a non-trivial task. With the creation of the VRPN library by Russell M. Taylor II at the University of North Carolina at Chapel Hill, the problem of integrating the many different devices was facilitated. VRPN provides a standard interface for all sorts of input devices. Despite VRPN's immense usefulness, the form in which device data is mapped and interpreted is still the same independently of the application. When running interactive research experiments, though, it is desirable to modify the input mappings and data interpretation of a device to identify potential optimizations in an application's input interface. With that in mind, the Remote Input library was designed and created. Its first version is already available and currently being tested. An introduction to its concept as well as a discussion on its usefulness, potential applications and foreseeable improvements will be the subject of this presentation.

More and more researchers are focusing on the management, querying and pattern mining of streaming data. The visualization of streaming data, however, is still a very new topic. Streaming data is very similar to time-series data since each datapoint has a time dimension. Although the latter has been well studied in the areas of information visualization, a key characteristic of streaming data, unbound and large-scale input, is rarely investigated. Moreover, the recent techniques on time-series data mainly focus on univariate data, and they seldom convey multidimensional relationship within the time-series data, which is an important requirement in the real application area. Therefore, it is necessary to develop appropriate techniques for streaming data instead of directly applying time-series visualization techniques to it.

In the dissertation proposal, I will discuss my plan to construct a framework for visualizing multivariate data streams, where each datapoint is a multi-dimensional vector. Two subtasks include streaming data visualization and data stream history visualization. In the first subtask, I plan to introduce some streaming data management strategies, including sliding window and load shedding techniques, into the traditional multivariate visualization. Old data or unimportant data will be discarded from the visualizations in terms of rules based on these strategies, so users can focus on most recent data as well as have a reasonable clutter amount. For the second subtask, I will design two types of visualizations to enable users to observe data trends and how data patter change. One's basic idea is to list multiple views corresponding to different time periods. The other uses streamlines to convey the change of cluster. After giving the overview of the above potential visualization techniques, the user studies to investigate the effectiveness and efficiency of visualizations will be described.

Augmented reality (AR) is the mixing of computer-generated stimuli with real-world stimuli. In this paper, we present results from a controlled, empirical study comparing three ways of delivering spatialized audio for AR applications: a speaker array, headphones, and a bone-conduction headset. Analogous to optical-see-through AR in the visual domain, Hear-Through AR allows users to receive computer-generated audio using the bone-conduction headset, and real-world audio using their unoccluded ears. Our results show that subjects achieved the best accuracy using a speaker array physically located around the listener when stationary sounds were played, but that there was no difference in accuracy between the speaker array and the bone-conduction device for sounds that were moving, and that both devices outperformed standard headphones for moving sounds. Subjective comments by subjects following the experiment support this performance data.

Paper Link

Single Photon Emission Computed Tomography (SPECT) is a medical imaging modality which allows us to visualize functional information about a patient’s specific organ or body systems. During 20-30 minutes scan, patients may move, causing misalignment, degrading image quality and potentially leading to errors in diagnosis. To improve the quality of SPECT imaging, we have to correct the motion. Previous work has modeled and corrected rigid body motion and periodic respiratory motion. Body bending and body twisting are kinds of non-rigid and non-periodic patient motion which may also occur during SPECT imaging. To correct this motion, we propose a deformation model to describe the deformation. We also introduce a way to extract the parameters of the model. Then we demonstrate its application within a modified iterative reconstruction algorithm. Several experiments will be designed to illustrate the correctness of the model and the modified reconstruction algorithm. Finally, an experiment with SPECT scans will be proposed to show the contribution of our deformation correction for SPECT imaging.

Recent advances in technology have enabled smaller, cheaper digital cameras with increased image resolution. However, despite significant improvements in camera components such as LCD displays and Integrated Circuits, the sequence and order of image processing steps used to create digital images has changed very little over the years. Also, beyond a few simple adjustments, such as color, contrast and gamma correction, the camera user cannot alter how the digital camera generates an image. For example, if a camera user wanted to apply cartoon-like effects to a captured image, they would have to post-process captured images using third party software such as Adobe Photoshop. To give the camera user more control over the way pictures are taken, we propose replacing the current static architecture of picture and video cameras with a new camera architecture that is made up of on-camera software programmable components. Our futuristic camera allows users to upload small programs, called Camera Shaders, which implement algorithms to manipulate images while being captured on the camera. Using an efficient stream processor and user-supplied Camera Shaders the camera user can apply and visualize sophisticated effects on captured images in real-time. As a benefit, long turnaround times and expensive retakes of shots can be minimized; yielding savings in movie budgets and bringing sophisticated effects to the masses. Our new camera architecture has implications on a wide variety of fields such as cinematography, robotics, consumer electronics, and could fundamentally change the way pictures and videos are taken.

As handheld devices have become increasingly popular, powerful programmable graphics hardware for mobile and handheld devices has been deployed. While many resources on mobile devices are limited, the predominant problem for mobile devices is their limited battery power. Several techniques have been proposed to increase the energy efficiency of mobile applications to improve battery life. We propose a new application-level dynamic voltage and frequency scaling (DVFS) on Graphics Processing Units (GPU). In most cases, cues within the graphics application can be used to predict portions of a GPU that will be used or unused when the application is run. We partition the GPU in to six clock domains that can be clocked at different rates. Specifically, each domain it has its own voltage and frequency setting based on its predicted workload to save energy without reducing frame rates. We propose an API that can be used to control these multiple clock domains. Finally, we propose an algorithm for predicting the workload offered to our six clock domains by a given application to decide voltage and frequency settings.