Below is a picture of my wife and I on the Golden Gate Bridge.
I am interested in graphics algorithms that perform object identification and 3D scene reconstruction. More specifically, I want to take an approach that utilizes a hybrid of human and computer, wherein humans do things that are easy for humans and computers do things that are easy for computers.
As an example, humans can easily look at a photograph and click on various objects, labeleing them as people, buildings, cars, trees, etc. Identifying flat ground, slopes, walls, and roofs is also easy for humans. In addition, material identification (metal, plastic, wood, etc.) is simple for a human. All previously mentioned things are difficult to identify with image processing algorithms.
Once a computer has some information from a human, interpolating and extrapolating as necessary to reubild 3D scene information should be far easier for a computer than a human. I think this could make for an interesting research area.
This paper talks a bit about constructing a 3D coordinate system, based on placement of a "matrix" on an object. The matrix is a simple black and white printed pattern that's attached to an object, much like a QR code.
This is interesting because it shows how the human needs to supply just a little but of information about the scene / image, then the computer can figure out where things are.
This paper talks about how humans perceive objects, then relates this to 2D image processing. The claim is that what humans are recognizing in imagery is "can be derived from contrasts of five readily detectable properties of edges in a two-dimensional image".
This is a more recent paper than the others, but it at least talked about putting an object-detection algorithm to work on a construction site and obtaining "promising results".
Where I might start on my research is to use photos of simple, primitive buildings, since these are often a combination of simple cube-like shapes. What I hadn't considered before seeing this article, is to use photos of buildings still under construction, since such buildings may be even easier to identify.
The paper's specific approach distinguished mobile equipment from workers, and wasn't as much focused on the architecture. But the findings still may be of interest for me.
(Do PhD students really have time for "hobbies"?)
My wife and I like eating good food from a variety of cuisines. Thai food and Italian food are among the favorites. If you're ever in Redwood City, try Arya Steakhouse. They have the best filet mignon I've ever tasted in my life.
I used to write a lot code as a hobby. Most of my projects were simple explorations for learning purposes, such as implementing a web server from the TCP socket level in C# and C++. Now, between work, school, and married life, I don't really have time for these projects anymore.