I've broken this into two categories. The first is personal interest, which is the research that I would do if paid to work on the topics of my choice. The second is professional, which is research topics that I've become (or will become) familiar due to projects that I've been involved in.
I have been asked if I find it restrictive and/or frustrating to be assigned projects/research issues, especially those that are not of personal interest. I have generally found it valuable to work with concepts outside of my preferred research niches; it leads me to identifying new possibilities, both in my preferred areas and in the potential synergies. And, as in the case of steganography/steganalysis, I find new areas of interest. In other words, no, it hasn't been restrictive.
Machine learning, in particular, nonparametric classification, has been my primary preoccupation over the past several years. However, I arrived at the conclusion, as many others have in machine learning, that eventually, to get a benefit from the techniques, you must apply them to one or more domains. Over the past year or two, I've become interested in two domains, namely 'Health Prediction' and 'Traffic Safety'. In the former, using data such as MRIs, PET, Clinical evaluations, and so forth, the goal has been to determine if a patient has dementia, is in a normal state, or is neither. In Traffic Safety, I, along with a few others, are attempting to predict the likelihood of accidents occurring and/or if an accident occurs, what is the likely results. In the past, I've also worked on equipment prognostics (predicting a failure before it occurs and hopefully preventing it) and image steganography (hiding information in the images)/image steganalysis (determining if information has been hidden).
My primary duty is to work with members of the Center for Advanced Computer Studies (CACS) to transfer their technology into commercial products. Since 2002, I've become familiar (to a greater or lesser extent) with
The information retrieval work was interesting, although I occasionally tripped (and still do) when encountered with the occasional differences between the IR perspective and the ML perspective. The image content extraction was a byproduct of the IR projects I've been involved; namely image retrieval.
Starting in July 2005, the projects in which I had been involved were closed. At that time, I was assigned to a new project, whose core areas represent a large departure from the areas in which I've been involved. The shorthand for the project was Photogrammetry. Essentially, the project was looking at how to automatically generate 3D scenes from a series of 2D images. Most of my involvement dealt with creating libraries, integrating third party packages and working on matching algorithms to find high quality point coorespondances.
That project closed in June 2007. At present, I'm involved in three primary projects. The first is essentially software engineering/development oriented. The goal is convert designated open source software, written to run upon a single workstation, to operate within a cluster environment; in other words, the idea is to distribute the functionality. The second project is video-related. In that project, a team composed of members with backgrounds in safety/traffic analysis and image/video processing, are seeking to analyze image and video fata collected from the driver's perspective and to extract information pertinent to the roadyway, including the road surface, the shoulder areas, guardrails, traddic control devices, and other roadside elements. The information is then to be rendered into 3D environements. The third project deals with Patient Health Prediction; this was a recent project that started in March 2008. Technically, it has ended, although follow-up funding is being pursued.
Thus, the project areas of interest are: