Interested in an Independent Study Project this or next semester? Do you like Computer Graphics, Computer Vision, Robotics, or playing with gadgets?

I have several potential projects for interested students. Some topics might be more than one-semester of work, but we could carve out a portion to fit into one semester. Depending on the project you choose, formal graphics background is not necessarily required. Below are several topics for projects. You can also stop by my office (CS154) and read a more complete list of topics from my office door or ask me in person or by email (aliaga@cs.purdue.edu).

Several topics:

*** Incremental Capture Planning ***

What path would you follow through a building to capture all surfaces in the least amount of time? What would an incremental capture-path look like? If you are designing a path so that you can capture the building, how can you design the path without knowing the building first (e.g., chicken-and-egg problem). In general, when capturing large environments, one important aspect is planning where, what, and how to capture the surfaces and objects of the environment. This problem is similar to the classical Art-Gallery problem (i.e., where do you place guards in a museum so that all surfaces, or paintings, are always under surveillance). In the 3D capture context, the question is where do you (automatically) drive around a semi-autonomous robot to acquire 3D information of an environment?

*** Semi-Autonomous Robotic Image-Capture Device ***

Do you like to play with robots, computers, and cameras? Think they are cool? Maybe you'd like to help build a robot-like device to catch a dense collection of images of a large environment. See pictures to right for an early prototype system.

*** Large-scale pose estimation system ***

Like knowing where you are? Do you also like GPS, lasers, and light beacons? How would you capture images through a large interior environment and know exactly where you are (without having to install a large custom installation)? You could use landmarks, you can use a laser-positioning system, or some other method. If you have ideas or interest in large-scale camera pose estimation, let me know!

*** Extracting Geometry (and BRDF) Information from a Dense Sea of Images ***

If you were given a large and dense collection of images of an environment, how would you extract geometry information. Classical stereo-reconstruction uses a few images to triangulate the position of certain objects or features. What you do if you have 20, 2000, or 20000 images of the same feature? Does this make the problem harder? Easier? More robust? More fun? What if people are moving throughout the images? Is this a large-dimensional search and optimization problem? Or maybe an application of level-set methods? Or maybe something totally new?

Any questions, just ask!