Depth estimation through iterative filtering uses a filter that refines depth maps from previous iterations. Its basis is using a bilateral filter using two penalty descriptors, the color of a region and the similarity of two disparity-matched regions. This research is focused on finding improvements of this method, mostly using a hierarchical approach. We use previously conducted experiments to compare the improved method with the original method, as well well as comparing the new method with results from other methods.
Link to full paper (PDF, 475k)
Link to presentation (PDF, 2.5M)
VDOWall is a realtime video compositing tool I am working on. It can be used as a VJ tool or as a tool for displaying certain presentations and such. We also use it in combination with our Photo Booth (details later).
More examples can be found here.
One of the projects we are working on internally at Unc Inc is video recognition. We want to be able to detect certain patterns in live video streams. I was already able to do some testing by using the OpenCV library. This will be worked out later.
All the stuff that can not be easily categorized (or when I am too lazy to do so) goes here.