Skip to main content
Held 27 April 2011 in 011 Sitterson Hall

H264Katana: Slicing and dicing H.264-compressed video streams

Max Beckman-Harned, supervised by Ketan Mayer-Patel

The field of video compression has yielded many algorithms and techniques to encode small files while maintaining a high level of quality, and H.264, today’s most common video compression codec, uses many of them. The current ecosystem of H.264 decoders focus on speed, taking as many shortcuts as possible to get from bitstream to decoded picture as quickly as possible. This talk will give an overview of compression algorithms, H.264, and the design and implementation of the H264Katana toolkit. H264Katana is designed to unravel H.264 bitstreams incrementally, providing an approachable and extendable framework for examining the elements in an H.264 bitstream, with the end goal of being able to perform analysis and editing of arbitrary elements in the bitstream as a basis for experimental study.

Max Beckman-Harned is a senior from Cary, N.C. Ever since his first words (“on” and “off”), he’s been destined to study Computer Science. Highlights of his time at UNC include four years of the ACM Programming Competition, his work with IT policy in Student Government, his job with ResNET doing tech support, and his Systems courses. He is the inaugural winner of the departmental Weiss Award for Outstanding Achievement in Computer Science. After graduation, he plans to join Microsoft in Mountain View, Calif., as a Software Development Engineer in Test for Hotmail.

Optimizing Ancestry Inference for Complex Pedigrees

Abhishek Sarkar, supervised by Wei Wang

Ancestry inference and the related problem of haplotype inference are essential for many genetic applications such as genotype imputation and linkage mapping. However, traditional inference algorithms based on Hidden Markov Models suffer from exponential state spaces, making them infeasible for many realistic datasets. In particular, the complex pedigrees of model organisms present challenges for this class of algorithms. Here, we present algorithmic optimizations to the Lander–Green algorithm which make the analysis tractable for model organisms. The optimizations do not compromise the accuracy of the inference but scale much better with pedigree size. We provide an implementation of this optimized algorithm at .

Abhishek Sarkar is a senior Computer Science major from Cary, NC. His main research interests are in applying computational algorithms and tools to solve biological problems. Currently, he is working on techniques for modeling genetic recombination and associating genetic features with physical traits. Abhishek is particularly interested in advancing understanding of the genetic basis of complex traits (arising from the interaction of many factors) such as susceptibility to disease. He plans to work towards this end by pursuing a Ph.D. at MIT.

Low-Cost Robot Platform for Medical Robotics Research

Pavel Chtcheprov, supervised by Ron Alterovitz

This talk presents a low-cost robot platform that mimics the capabilities of currently used laparoscopic surgical robots and that can be used in a research setting to evaluate algorithms that would enable the automatic performance of surgical tasks. Industrial robots are able to provide speed, precision, accuracy, the ability to perform complex tasks, and most importantly, automation of these complex tasks. Current medical robots, on the other hand, must be tele-operated in part due to the variability of the surgical scene. Ongoing research is investigating computational methods to automate surgical procedures and use a vision system to sense changes in the dynamic environment in real time. The goal of this project is to design and build a low-cost robot test bed to evaluate new motion planning and 3D computer vision algorithms for robot-assisted minimally invasive surgery. The low cost robot created in this work consists of an articulated robotic arm with six degrees of freedom, mimics a laparoscopic instrument with a single pivot point at the theoretical skin interface, and features an interchangeable end effector that supports both camera and laparoscopic tool attachments. The robot’s kinematics and inverse kinematics were derived, and the robot was interfaced with a computer to enable control using standard computer languages. The new low cost robot platform provides an actual robotic system on which to test and evaluate new motion planning algorithms, enabling experiments that take into account mechanical uncertainties. The system was evaluated using two experiments: (1) an object grasping task involving the gripper attachment, and (2) an accuracy experiment involving a pencil attachment to visualize the path of the end effector. This low-cost robot that mimics the capabilities of current laparoscopic surgical robots has the potential to further research in the area of automated surgical tasks.

Pavel Chtcheprov will be graduating with Highest Honors in Biomedical Engineering from UNC-Chapel Hill in spring 2011. As an undergraduate student, he conducted research in computational and experimental fluid dynamics, tissue engineering and synthetic biology, and medical robotics. He has contributed to one publication in a peer reviewed journal and to four conference abstracts, was awarded two NIH Integrated Biomedical Research Training Program (IBRTP) Fellowships, and won a Sarah Steele Danhoff Undergraduate Research Award. He will be joining the UNC-NCSU Biomedical Engineering Program for graduate school in fall of 2011.

Hookt on Fon-iks

Austin Matthews, supervised by Fabian Monrose

In this work I seek to develop methods for reconstructing English sentences from an unsegmented string of English phonemes. Unlike work in speech-recognition, I interpret sentences given only a phonetic transcription of speakers’ real-world pronunciations. In particular, I will present an algorithm for segmenting phoneme strings into words and an algorithm for matching pronunciations with English words, both resistant to the affects of input noise, speaker dialect, or non-standard pronunciation.

Austin Matthews was born and raised in Greensboro, N.C., and attended the North Carolina School of Science and Mathematics before coming to UNC. Here at UNC he studies Mathematics, Computer Science, and Linguistics, and holds a particular interest in the intersection of the three. He will be pursing these interests next year at Carnegie Mellon University’s Language Technology Institute. In his free time he enjoys game programming and playing chess.

The Bathysphere: Motion Capture as Art

Caitlyn Losee, supervised by Greg Welch

The University of North Carolina at Chapel Hill’s Institute for the Arts and Humanities works to promote collaboration across disciplines. In February 2010, this institute coordinated the first CHAT Festival (Collaborations: Humanities, Arts and Technology), which involved several universities in the Triangle area; UNC-CH hosted the event. We created an engaging art installation for the festival where audience members stepped into a small building and entered an underwater world. The audience stood on the sandy sea floor, ocean stretching out into the distance around them. As they moved the physical objects in the space, sea creatures mirrored their motions in the underwater environment. This system employed motion capture technology to continuously measure and record the three dimensional position and orientation of the physical objects, a game development tool to move the models and render the environment, and multiple projectors to create the large-scale immersive environment. In this talk I’ll give a technical overview of the system including the approaches used for display, motion capture, and graphics.

Caitlyn Losee is a Chapel Hill native finishing her fourth year of the new five year BS-MS program in Computer Science. She completed a minor in Math. Last summer she was an Extreme Blue intern at IBM’s Almaden Research Center in San Jose. This summer she will be interning with IBM in Research Triangle Park.

Modeling Essential Tremors (ET)

Krishna Kollu, supervised by Greg Welch

In this talk we describe the motivation for, and current status of, the development of mathematical models for neurons engaged in the cyclical behavior that causes a neurological disorder called essential tremors. The results of this modeling will be used for teaching medical students about the neurological behavior, and to develop simulations that can eventually be used to develop automated closed-loop system for controlling deep brain stimulation units.

Krishna Kollu is a junior, majoring in computer science. For the past year he has been working with Prof. Greg Welch (Computer Science), Dr. Richard Murrow (UNC Neurology), and more recently Prof. Oleg Favorov (Biomedical Engineering).

Browser Based Sonic Zoom: Using Javascript, HTML5, and Spatial Sound as Enabling Technology

Cameron Swaim, supervised by Gary Bishop

In public schools, children spend time in the computer lab reading interactive storybooks, playing games, and using educational programs. For children who are blind or visually impaired, this leaves them with very few options, possibly nothing. Sonic Zoom was written in Spring 2004 by a group of students working on a semester project to help alleviate this problem, and has become a major hit in the blind and visually impaired community. Sonic Zoom was developed as a Microsoft Windows and DirectX application. With the advancements in technology since that time in HTML5 and Javascript, remaking the classic and well loved game for the browser would allow easier distribution and widespread access. HTML5, the Dojo Javascript Framework, and sound editing applications were used to remake Sonic Zoom for the browser. The game works very similarly to the original Sonic Zoom, using all of the same sounds and controls, but is now available to anyone with an HTML5 and Javascript enabled browser.

Cameron Swaim is a senior Computer Science major from Sherrills Ford, N.C. His research under Professor Gary Bishop focuses on enabling technology and emerging web technologies. Currently he is working for Adzerk, a startup in Durham at the American Tobacco Campus that specializes in ad server technology.

Effect of Transient Cross Traffic on the RAPID Congestion Control Protocol

Rebecca Lovewell, supervised by Jasleen Kaur

The RAPID packet-scale congestion control protocol is a new paradigm that allows end-to-end transport protocols to scale to ultra-high speed networks. While most existing protocols can attain a single-stream throughput of no more than a few Gigabits-per-second, this paradigm promises to scale up to Terabit-and-higher speeds.  It does so by analyzing how finely-controlled inter-packet gaps change in the network. However, RAPID faces several open challenges, including sensitivity to noise in the end-to-end delays experienced by packets. In particular, RAPID is especially responsive to transient cross traffic at bottleneck links within a network. In this project, we conduct an experimental analysis to quantify the extent to which such transient cross traffic can impact the RAPID throughput. Preliminary results help form guidelines on how to configure the protocol parameters.

Rebecca Lovewell is a senior computer science student from Winston-Salem, N.C.  She has worked with Dr. Jasleen Kaur since January 2011.