The following is a list of potential projects that are available in Summer 2009. The text was provided by the corresponding mentors. Currently, each of the projects involves two students. Some changes in the nature of the projects may occur, based on the faculty/student interest and background. Background literature and computer application training will be provided for all the projects. For future information on any of them, feel free to contact the site coordinator, Dr. Stefan Robila or any of the mentors listed below.
Autonomous Vehicle Research (Mentor: Peng)
The autonomous vehicle will have the ability to drive from one position to another on a GPS map. The system will have many positions entered into it, describing point to point directions. GPS in conjunction with an inertial navigation system (INS) will then verify where the vehicle is located in respect to the next desired longitudinal and latitudinal position. A path programming system will choose the best (smoothest and shortest) path to travel from position to position. There will be two laser scanners produced by SICK that will be mounted on the top of the vehicle. Standard practice is to leave the units stationary. Instead we will turn the units horizontally and oscillate them such that there internal vertical scan now has an external horizontal motion. This will allow us to view the front of the vehicle fully whether the vehicle is moved or stopped. We now have a periodically upgrading 3 dimensional view of the front of the vehicle. This data can then be analyzed to determine what is and is not an obstacle, which can then be added to the GPS mapping system so that in the event of a confirmed obstacle, the path programming system can adjust around it. This type of autonomous system can make driving easier for handicapped persons and increase general roadway safety. The first prototype will be a small scale golf cart, allowing it to be easily tested on campus grounds. In this project, students will examine ways to interpret signals returned by the laser scanners and combine the signals with position data returned by GPS and INS to program a path for the vehicle using a B-spline path algorithm. It is challenging because it involves interpreting a massive amount of information from the laser scanners and then casting it onto a GPS and INS mapping system, as well as controlling a drive-by-wire system with great precision to minimize course error. The vehicle and its apparatus provide a very flexible foundation to work with. This, coupled with programming done in C++ or JAVA, allows for many years of research and improvements to be made.
Efficient Processing of Hyperspectral Imaging (Mentor: Robila)
Student research projects in hyperspectral imaging involve design, implementation, and experimentation with different software modules and hyperspectral sensors. In many science fields, the sensed data are collected as images (spectral bands), with each image corresponding to intervals of wavelengths. A collection of spectral bands over several wavelength intervals for the same scene is called a multispectral image. When the spectral measurement is performed using hundreds of narrow contiguous wavelength intervals, the resulting data are called hyperspectral images. Hyperspectral images provide a considerable amount of information allowing for detection of slight differences in materials otherwise not distinguishable with ‘traditional’ imaging techniques. Such data are increasingly used in a wide array of sciences including agriculture, geology, biology, medical imaging, etc. A current area of interest is grid technology used in hyperspectral image processing. In this project, you will learn about the nature of the data as well as use available grid technologies to process the images. An additional area is the inclusion of spectral data in face recognition.
Efficient Data Visualization for Data Beyond the Human Visible Spectrum (Mentor: Gutierrez)
Visualization is defined as the graphical presentation of information, with the goal of providing the viewer with a qualitative understanding of the information content, since the human visual system has very good advanced information processing abilities. One possible direction you will be working on will be to select a hyperspectral data set and investigate visualization approaches that would enhance the understanding of the statistical properties of the data. In previous projects, REU students developed a weighted composite scheme for data visualization that was simple, efficient, and outperformed other more complex techniques. For this year, the students will use again relevant parts of the non-visible spectrum. A promising approach is the use of wavelet transforms on these areas and the visible spectrum. We expect to compare the results with the ones from previous years.Comparing Wavelet Based and Feature Based Approaches for Image Processing (Mentor: Varde)
It is found in the literature that images are often processed using approaches that fall in the general category of either wavelets or features. Wavelet based methods use a mathematical function called a wavelet transform to split the original image into multiple scale components, studying each component with a resolution corresponding to its scale in order to process the image. Feature based methods as the name implies consider various features of the image such as color, grayscale, depth, inter-particle distance along domain-specific features if needed, performing image processing typically by using a weighted sum of these features. The goals of the image processing can be multifold catering to techniques such as similarity search and clustering, with a range of applications such as simulation tools and expert systems. In this project we aim to compare these two approaches for image processing, namely, wavelet based and feature based. We consider accuracy and efficiency of processing as the most important criteria for comparison along with other factors as deemed necessary in specific techniques and applications. Finally, we also consider an ensemble of wavelets and features to develop more enhanced image processing methods. The images we use are mostly from scientific domains.