Browsing by Author "Boyd, Jeffrey"
Now showing 1 - 14 of 14
Results Per Page
Sort Options
- ItemOpen AccessAutomatic object tracking(2006) Zhang, Yonghua Michael; Boyd, Jeffrey
- ItemOpen AccessEnergy-Efficient Workload Placement with Bounded Slowdown in Disaggregated Datacenters(2023-12-08) Sefati, Amirhossein; Ghaderi, Majid; Krishnamurthy, Diwakar; Boyd, Jeffrey; Ghaderi, MajidDisaggregated Data Center (DDC) is a modern datacenter architecture that decouples hardware resources from monolithic servers into pools of resources that can be dynamically composed to match diverse workload requirements. While disaggregation improves resource utilization, it could negatively impact workload slowdown due to the latency of accessing disaggregated resources over the datacenter network. To this end, we consider CPU and memory disaggregation and conduct measurements to experimentally profile several popular datacenter workloads in order to characterize the impact of disaggregation on workload execution slowdown. We then develop a workload placement algorithm, called Iterative Rounding-based Placement ( IRoP), that given a set of workloads, determines where to place each workload (i.e., on which CPU) and how much local and remote memory is allocated to it. The key insight in designing IRoP is that the impact of remote memory latency on slowdown can be substantially masked by assigning workloads to higher-performing CPUs, albeit at the cost of higher power consumption. As such, IRoP aims to find a workload placement that minimizes the DDC power consumption while respecting a bounded slowdown for each workload. We provide extensive simulation results to demonstrate the flexibility of IRoP in providing a wide range of trade-offs between power consumption and workload slowdown. We also compare IRoP with several existing baselines. Our results indicate that IRoP can reduce power consumption and slowdown in the considered scenarios by up to 8% and 12%, respectively.
- ItemOpen AccessEvaluation of Complex Surveillance Systems for Emergent Vulnerability(2010-09-08T18:38:20Z) Thornton, Christopher; Cohen, Ori; Denzinger, Jorg; Boyd, JeffreyThe current paradigm for testing tracking and surveillance systems is to identify representative metrics for system components, then optimize the performance of that metric against test data. The assumption is that optimization of individual components will optimize the surveillance system as a whole. However, while optimizing components is a necessary step to improve systems, it is not sufficient to address vulnerabilities that emerge in a large system with many components. A large surveillance system will have many cameras and other sensors. In some cases, to cover more area, the cameras and sensors may be mobile. Coverage is unlikely to be complete in all areas at all times, so sensor allocation will follow some policy. The combination of sensors, sensor properties, mobility and policy can result in a system that is vulnerable in ways that are difficult to predict. We present a method to model and predict emergent vulnerabilities in a complex surveillance system. To demonstrate the method, we apply it to a downscaled physical surveillance system that uses multiple stationary and mobile camera platforms to monitor and defend against intrusions. Our method finds two vulnerabilities in the system in simulation, one of which we demonstrate with the physical system.
- ItemOpen AccessExploring Higher Dimensional Photography Using Programmable Cameras(2013-07-10) Willson, Christopher; Boyd, Jeffrey; Denzinger, JörgConsumer-grade cameras can be programmed by technical users with an interest in photography, but currently the potential is still being investigated. We explore some of the capabilities available using programmable cameras through the example of capturing what we term a four-dimensional (4D) volume of image data. Having access to a 4D volume of image data allows for the creation of photographs that are visually interesting, but also technically challenging to work with. We deal with the inherent complexity by building tools using computer vision and image processing techniques both at the computer desktop software level as well as inside the camera itself. Our intention is to assist artists and photographers to manage this inherent complexity. The assistance we provide is evaluated by some artists and photographers through a user study. Now that programmable cameras have become available, our work shows how embedding image processing inside a consumer camera along with other computer vision tools can facilitate photographers’ creative possibilities.
- ItemOpen AccessHuman observation and computer analysis of human running gaits(2005) Kendrick, Nathan; Boyd, Jeffrey
- ItemOpen AccessMotion swarms: video motion interaction for large audiences(2008) Nguyen, Quoc; Boyd, Jeffrey
- ItemOpen AccessMUSE: a Music Sandbox Environment for Novices(2017) Popa, Iulius; Boyd, Jeffrey; Eagle, David; Boyd, Jeffrey; Eagle, David; Radford, Laurie; Manzara, LeonardCollaborative musical interfaces for novices allow people with limited to no musical education access to a “walk-up and play” group musical experience. However, research shows that the ease-of-use of an interface tends to be inversely proportional with its creative affordances. My research aims at increasing the level of creative controls in interfaces for novices without compromising the “walk-up and play” characteristic or negatively affecting the perceived quality of the musical output. This thesis proposes a new design paradigm – a “game of music” and presents the design, implementation, and evaluation of MUSE – a real time, collaborative musical interface for novices. Participants in user studies of MUSE found learning and using the interface easy, described the music as pleasant, and reported having creative control over the music output. Further research could explore this new paradigm and its potential to maximize creative affordances when designing new collaborative musical interactions for novices.
- ItemOpen AccessNoise Reduction And Information Extraction Of Dual-Energy Computed Tomography Images(2016) Simon Maia, Rafael; Jacob, Christian; Mitchell, Joseph Ross; Boyd, Jeffrey; Cunningham, Ian; Costa Sousa, Mario; Frayne, RichardWith every new generation of computed tomography machinery, big improvements in terms of image quality and acquisition speed were achieved. Nonetheless, a persistent feature of the CT remained: the single energy acquisition of images, which results in images where materials of similar density appear with similar intensity, generating an undesired degree of uncertainty that required other kinds of examinations to be solved. However, it has been known since its invention that the instantaneous acquisition of two or more energies would provide images with better tissue discrimination capabilities and reduce this problem. Nonetheless, it was only in the last 10 years that CT technology became advanced enough to simultaneously acquire images in dual-energy mode. However, it is necessary to keep the radiation dose to the patient equivalent to a single energy CT image, which results in images that are affected by noise and that need especial algorithms to improve the image quality. A particular feature that has also been know since the invention of dual energy CT is the characteristic negative correlation of its material density information discovered by Kalender et al. In this work we developed two algorithms that takes advantage of that discovery. We first created an algorithm using a joint anisotropic diffusion that reduced the amount of noise and improved image quality. Finally, we extended this first algorithm by using an adaptive Wiener filter that better approximated the true mean value of each region and drastically improved image quality even in images that were deeply affected by noise. The proposed techniques were tested in a quantitative way in simulated, real phantom and real patient images to show the improvement in image quality while preserving image information. Finally, we investigated these noise corrected images in the perspective of information extraction, using a modified multi-material decomposition algorithm to obtain a classification of pixel in term of tissue type.
- ItemOpen AccessSensing Rhythm: Synchronizing Auditory Feedback and Movement(2016) Godbout, Andrew Joseph; Boyd, Jeffrey; Samavati, Faramarz Famil; Katz, LarryThis thesis explores auditory feedback, designed around the ability of people to entrain their movements with a rhythm. Phase denotes the progression through a movement and thus tracking phase tracks progression through a movement. We show how to track phase using sensors and develop a novel visual synchronization system that works with RGB-depth cameras. We use phase tracking to produce auditory feedback that is synchronized with a user. Applications to walking, running, cycling and speed skating demonstrate the methods and thirteen speed skating athletes test the robustness and provide qualitative feedback of two of our synchronous auditory feedback systems.
- ItemOpen AccessSonification of gait(2009) Sadikali, Akil; Boyd, Jeffrey
- ItemOpen AccessTranslating technology to clinical practice: Predicting how knee osteoarthritis patients will respond to an exercise intervention(2017) Kobsar, Dylan; Ferber, Reed; Hettinga, Blayne; Boyd, JeffreyMuscle strengthening exercises consistently demonstrate improvements in the pain and function of adults with knee osteoarthritis, but individual response rates can vary greatly. Identifying individuals who are more likely to respond is important in developing more efficient rehabilitation programs for knee osteoarthritis. Therefore, the overall goal of this thesis was to identify responders to exercise with a conventional motion capture system and translate these findings into a clinically accessible wearable sensor system. It was found that a conventional motion capture system, in combination with patient-reported outcome measures (e.g., function) collected at the baseline of an exercise intervention can successfully predict responders to treatment with greater than 85% accuracy (chapter three). To translate these findings to the clinical setting, more accessible wearable sensors (e.g., accelerometers) were examined in the remaining chapters. Chapter four found that while a single sensor at the lower back could subgroup some gait patterns, it was not sensitive enough to separate other, more similar, gait patterns. Therefore, the reliability of using multiple wearable sensors was examined in chapter five. The lower back, thigh, shank, and foot were all found to be reliable sensor locations for gait analysis and therefore suitable in the final study as potential predictors of response. Finally, chapter six found that a unique combination of wearable sensor data and patient reported outcome measures could successfully identify responders to an exercise intervention with similar accuracy to the conventional motion capture system. Further, the best limited set of sensors included only the back and thigh. Therefore, these findings suggest the potential development of a simplified two sensor system that can provide clinicians with an efficient and relatively unobtrusive way to use to optimize treatment.
- ItemOpen AccessUtilizing Natural Language Processing to Produce and Present Film: An Exploratory Analysis(2023-01-06) Jorgensen, Kyle; Wang, Mea; Boyd, Jeffrey; Ovens, KatieThe Subscription Video-On-Demand (SVOD) market is highly competitive, with companies constantly vying for market share. Inspired by the natural language processing (NLP) technologies that have allowed virtual assistants to flourish, we investigate how NLP could help in film production and film presentation and look for novel ways in which SVOD digital media companies could benefit. In particular, on the film presentation side, we propose a novel context-aware question-answering system that would be integrated with SVOD content to allow viewers of video to ask context-sensitive questions about the film they are watching simply by pausing. On the film production side, we investigate the NLP performance of using natural text as an input in automatic character extraction (also known as co-reference resolution). We present results and prototypes that show that NLP has meaningful use cases, both in film production and film presentation.
- ItemOpen AccessVision-Based Stabilization for Fixed-Wing Flight(2011-12-05T17:13:55Z) Boyd, Jeffrey; Thornton, ChrisVision and flight are closely linked, leading to a longstanding interest in how the two are connected. Past research has proposed models for vision in control for biological examples of flight, and there has been recent interest in the use of vision for low-level control of small robotic aircraft such as quad-rotor helicopters. In the work presented here, we show a system for stabilization of a small, fixedwing aircraft in the yaw axis using estimates of parametric optical flow obtained by registration of consecutive video images from a camera mounted on the aircraft. Estimates of angular velocity from the registration replace the values that would otherwise come from a gyro in a conventional stabilization system. No markers or special targets are required – just an environment with enough visual variation to enable image registration. We demonstrate the system in flight and show qualitatively the efficacy of the stabilization from external observation of the aircraft, and from the video acquired from the onboard camera. 1 Introduction
- ItemOpen AccessVisualizations on a Web-Based View-Aware Digital Earth(2017) Sherlock, Mark; Samavati, Faramarz; Samavati, Faramarz; Boyd, Jeffrey; Wang, XinA Discrete Global Grid System (DGGS) is a powerful tool for creating the discrete reference models that support geospatial dataset integration, organization, processing, and visualization in a Digital Earth (DE) application. However, the growing size and scale of geospatial datasets presents significant obstacles to the interactivity and accessibility of geospatial visualizations. To address this challenge, we present a portable DGGS that runs in web browsers on a client device and efficiently communicates with a server-side DGGS. In our method, the client-side is responsible for triggering queries for missing data, managing the viewing area and rendering various styles and effects. The server is responsible for generating data representations for DGGS cells in response to queries from clients. Only the data needed for the current view and scale need to be processed, so the task of processing the datasets at their native resolution can be circumvented. The resulting system is capable of interactively displaying multiple simultaneous viewpoints, allowing for a multilevel focus+context visualization of the globe. We also present several real-time data styling techniques that are designed to work efficiently on both the client and server. These methods help make DE more accessible and informative than ever before.