Final Project

The final project can be done in groups of up to 2. It needs to include both implementation and experimentation. It can be turned in either as a webpage or as an electronic document (pdf or doc) file.

There is a mandatory project proposal/checkpoint that is due by April 29, which is to include your algorithm description (at least at a high level) and any roadblocks that you have come upon so far, and your experimental plan and what you have so far completed.

The project can be turned in anytime before March 9 at 5:30 pm.

The following options are potential final projects. Any ideas of this scope/flavor would be acceptable.

• Project 1. Tracking. Implement a particle filter *or* a Kalman filter *or* a mean-shift tracker in order to track people *or* cars in a scene. You will have to hard code some things for your tracker
• Project 2. Attach the Kitten CAPTCHA
  1. Generate or collect a reasonably large collection of examples of the kitten captcha test, or use google or flickr to generate your own.
  2. Write a brief (one page) report of all publically available information about the captcha, the specific rules (what does the captcha look like? what does the user have to do to "pass" the captcha? How long does it take for the user to pass the test? if a computer guessed randomly, how likely would it be to get it right? how are the images in the captcha chosen --- is there a library of images? how large of a library? Is this captcha sensitive to a "dictionary attack"?).
  3. Attach this captcha using tools for object recognition to classify images as "cat" or "not cat". This may include, perhaps, Hough transforms or template filters to find the circular eyes and or triangular ears, or characterizing the distributions of colors or SIFT features in a training set of cats (vs. non-cats), and using nearest neighbor classification.
  4. Characterize your classification performance. What percentage success rate will you have if you apply this classification performance to the CAPTCHA?
  5. This project will be graded on the write-up and the reasonableness and implementation of the approaches that you tried and not the final success with which you attach the captcha.
• Project 3. Implement a "livewire" segmentation tool.
• Project 4. Position papers in Topics in Computer Vision. (individuals only) Write about a question such as those listed below. You must take a stand on one side of the issue, and defend it. I don't like to give strict constraints on page lengths and such, but your analysis needs to address the issue, illustrate the technical/computational challenge, and defend a claim. This is likely to take 6-20 pages. It can be turned in as a paper or a website.
  1. The DARPA grand challenge robot race was won by a team that made more use of vision than any of the other competitors. In what ways may automated vision algorithms affect every-day commuters in the next 20 years?
  2. What has been the greatest impact of any computer vision product in the last 50 years, in terms of man-hours/lives saved or disasters prevented? What is likely to be the biggest impact in the next 20 years, and why?
  3. Video surveillance is becoming more and more common in public areas, and vision algorithms can (sometimes) track, and recognize people. What are the privacy implications of this? Is it privacy invasive now, even though the face recognition (of images of people on a sidewalk) is quite poor and tracking is mostly limited to just one camera? If we had perfect recognition and perfect tracking of everywhere you go in public, is that privacy invasive? At what point is it privacy invasive? These questions can be address in an ethical, philosophical or legal perspective. Some examples of potential problems can be found here.