NSF Research Experiences for Undergraduates (REU)
Summer 2010 - Computer Vision and Multi-Agent Systems
Department of Computer Science, Utah State University

  Name:

  Brian Forbis

  Major:

  Computer Science

  Minor:

  Mathematics

  School:

  Willamette University

  Home Town:

  Los Angeles, California

About Me

I am a 3rd year Computer Science student going on to my 4th and final year at Willamette University in Salem, Oregon. After days of nagging from my CS professors at Willamette, I decided to apply for NSF funded REU's around the country. The REU here at Utah State University happened to be one of the programs that accepted me and I decided to attend this research position because I was most interested in object detection research.

This has been a real change of pace for me as far as summer work has gone. My summer jobs so far (a background actor in Hollywood and a sailing instructor) have not required college level education, so I wasn't sure how I would be able to cope with the amount of work. This did not end up to be too much of a problem though. Although I ended up programming in a language I had never heard of, we had 2 weeks of classes to help us learn all that we needed to know to complete our projects.

Logan is actually an interesting town once you get to know it. I spent most of my days rock climbing the courses in Logan canyon and going to the indoor rock climbing wall a few miles away from campus. There are also several interesting places to go hiking in the mountains.

Project

This summer, I worked on License Plate Recognition software to recognize license plates from images of vehicles in a database. I used an algorithm that would identify the plates using the knowledge that license plates have high contrast and show up really well on thresholded images. I then used geometric conditional classifiers and neural networks to return plates from hundreds of possible plate images returned from each image. Much of my time was spent creating and optimizing small feature vectors (sets of numerical attributes used to represent an image) for use in their system.

Results

I ended up getting an accuracy of 99% and a precision of 66% for detecting license plates within images. I feel that if I can work out a few kinks in my code and convert the code to a faster language than MATLAB, I will be able to implement this program using a live feed camera for inputs rather than a database. You can read more about this in my final presentation or final paper.

Feel free to write me at: