Skip to main content

Secondary navigation:

2010 Bachelors Theses

computer science

 

Author: Mike Betten
 
download pdf
Title: Near Real Time Human Pose Estimation using Locality Sensitive Hashing
Advisor: Hao Jiang

Abstract

This thesis aims to develop a data set of poses to be used with locality sensitive hashing in human pose estimation, and then to determine the potential for real or near real time use of locality sensitive hashing in human pose estimation with that data set. Background subtraction is used to obtain a silhouette of the human pose from a single frame of video. The pose is then matched against the data set of poses using locality sensitive hashing. Results indicate that a data set can be developed to be used with locality sensitive hashing to match human poses in video and that near real time results are possible. The application of locality sensitive hashing for human pose estimation in video facilitates further discussion of the uses for pose estimation in video as well as the uses of locality sensitive hashing in fast feature matching

Author: Ben Brown
 
download pdf
Title: The Universal Content Management System (UCMS)
Advisor: Bill Ames

Abstract

There are millions of content management systems out there in the world, each fitted with a fairly narrow purpose. These systems can range from very simple ones to very advanced ones, but what they all have in common is that they require a programmer (or even team of programmers) to build them.

What I have attempted to do is to create a universal content management system—one which allows for (ideally) any kind of feature or set of features to be created with as little dependence on programmers and on other people with any significant understanding of technology. In other words, this system generates whole features and specific pages for these features on its own, with the user only specifying what sort of fields and other content he wants on those pages. This is code that writes code.

In order to make this possible, I have looked at the most general and most common design patterns for content management systems—specifically database design patterns—and incorporated them into this system. What pattern or set of patterns we use is entirely contingent on what options the user chooses, but the user does not need to know more than basic database relationship concepts in order to use the system effectively (assuming, of course, that he take some time to familiarize himself with the system and the terminology it uses).

UCMS can generate blogs and simple forums, among other features. At this stage, it is powerful enough to be a simple—moderate content management system, in terms of complexity. This is because UCMS still has some limitations, as there are some general design patterns and issues that have not yet been covered by it. Additionally, it is limited only to PHP and MySQL based content management systems (as this is the language and database system, respectively, that it has been built on and for).

Author: Greg Epstein
 
download pdf
Title: Harnessing User Data to Improve Facebook Features
Advisor: Sergio Alvarez

Abstract

The recent explosion of online social networking through sites like Twitter, MySpace, Facebook has millions of users spending hours a day sort- ing through information on their friends, coworkers and other contacts. These networks also house massive amounts of user activity information that is often used for advertising purposes but can be utilized for other activities as well. Facebook, now the most popular in terms of registered users, active users and page rank, has a sparse offering of built-in filtering and predictive tools such as “suggesting a friend” or the “Top News” feed filter. However these basic tools seem to underutilize the information that Facebook stores on all of its users. This paper explores how to better use available Facebook data to create more useful tools to assist users in sorting through their activities on Facebook.

Author: Ryan Gadsby
 
download pdf
Title:Implementations of Robotic Swarm Networks using Low-Cost Hardware
Advisor: Robert Signorile

Abstract

From the surface of Mars, to terrestrial car factories, robots play an important role in many aspects of human life. However, the limitations of robotic hardware–power consumption, imprecise movement, and monetary cost chief among them ­ are often a massive obstacle to the implementation of many possible applications of this technology. Swarm robotics provides several methods of circumventing these roadblocks. Smaller, more primitive robots typically consume less power and cost less money to develop than larger, more intricate machines. However, by distributing tasks among multiple smaller robots, the same amount of work can be accomplished as if we designated the same task to a single rover. The problem of imprecise movement can be alleviated by using multiple robots to interpolate their estimated surroundings with existing data instead of relying on a single machine's interpretation of its environments. By using a large network consisting of low­cost robots, we can take a more cavalier approach to the use of robots of exploration and reconnaissance: since even an inexpensive sensor on an expensive machine might be irreplaceable and therefore limit the possible applications of a given robot, a member of a swarm can disappear without inconveniencing its brothers unduly.

This thesis examines several different swarm network configurations and their performance at tasks analogous to real­world applications of swarm technology. This is achieved using a low­cost modular robotic controller developed by LEGO Mindstorms known as the NXT, and installing a Java Virtual Machine on it to provide a hassle­free development platform. The benefits of swarm technology are fully explored in the swarm's implementation as their collective sensory input is used to form a “hive mind” accessible by any given member of the swarm. I focused on the task of searching for an object in the physical environment, and compared the time spent and general effectiveness of the swarm to a single bot equipped with more sensors doing the same task.

Author: T. J. Keemon
 
download pdf
Title: 3D Reconstruction of Human Motion Through Video
Advisor: Hao Jiang

Abstract

In this thesis, we study computer vision methods to capture 3D human movements in videos. The goal is to extract high quality movement from general videos using minimum human interaction. With a few mouse clicks on the body joints in each video frame, all the possible 3D body configurations are automatically constructed and their likelihoods are quantified using a prior trained with millions of exemplars from the CMU motion capturing database. The 3D body movement is optimized using dynamic programming using the pose hypotheses and the temporal smoothness constraints. The proposed method can be used for unconstrained motion capturing and our experiments show that it is efficient and accurate. We further study two applications based on the proposed motion capturing method. The first one is to animate characters using the motion captured from videos. The second one is for sports performance analysis. With the 3D movement information, we can measure body part speed, coordination, and various other parameters.

Author: Dan Szafir
 
download pdf
Title: Non-Invasive BCI through EEG: An Exploration of The Utilization of Electroencephalography to Create Thought-Based Brain-Computer Interfaces
Advisor: Robert Signorile

Abstract

It has long been known that as neurons fire within the brain they produce measurable electrical activity. Electroencephalography (EEG) is the measurement and recording of these electrical signals using sensors arrayed across the scalp. Though there is copious research in using EEG technology in the fields of neuroscience and cognitive psychology, it is only recently that the possibility of utilizing EEG measurements as inputs in the control of computers has emerged. The idea of Brain-Computer Interfaces (BCIs) which allow the control of devices using brain signals evolved from the realm of science fiction to simple devices that currently exist. BCIs naturally present themselves to many extremely useful applications including prosthetic devices, restoring or aiding in communication and hearing, military applications, video gaming and virtual reality, and robotic control, and have the possibility of significantly improving the quality of life of many disabled individuals. However, current BCIs suffer from many problems including inaccuracies, delays between thought, detection, and action, exorbitant costs, and invasive surgeries. The purpose of this research is to examine the Emotiv EPOC© System as a cost-effective gateway to non-invasive portable EEG measurements and utilize it to build a thought-based BCI to control the Parallax Scribbler® robot. This research furthers the analysis of the current pros and cons of EEG technology as it pertains to BCIs and offers a glimpse of the future potential capabilities of BCI systems..