Tuesday, 9 February 2010

Visit to TU Dortmund: Impressive Demos on Vision and Audio

After several tries we finally managed to travel to Dortmund (half an hour on the S-Train) to visit Gernot A. Fink's group at the Technical University Dortmund. Bastian Pfleging did with this group his master thesis before he joined us. The research focus of the group is on signal processing and computer vision. They also follow an experimental approach - building systems that work (which we saw in the demos). In their lab space they have setup a building (basically a house inside a house - impressive!).

I have learned about a new location technology based on passive infrared sensors. The idea is to pick heat emitted from people and combine the output from several sensors to localize the person. The technology is very simple, potentially cheap, and privacy preserving. Sometime back we thought of a project topic using thermal imaging (not really cheap or privacy preserving) for context-awarenes - but so far there was no student who wanted to do it. Perhaps we should try again to find a student.

The other demos were situated in a meeting room that is equipped with several cameras and microphones. It was interesting to see how robust several of the vision prototypes managed to track people in the room and to detect pointing actions. One basic mechanism the use to detect interesting regions in an image is saliency based on different features - and it works well.

The audio demo used two arrays of 8 microphones each; the arrays are nicely integrated in a ceiling panel. Using these signals they can calculate the energy that originates from a certain spatial region in the room. Looking at the complexity of the hardware and software for sound localization it appears not in the far future that this could become ubiquitous. We talked about the work James Scott did on sound localization (snipping on a light switch) - here is the reference [1].

The room is equipped with sensors, lights, switches and a UI panel that are linked over a commercial bus system (KNX). Sometime ago we had a bachelor project in Essen that looked at EnOcean (another home networking technology). We discussed how well these systems are positioned in comparison to web technologies.

I personally think medium term we will move - at least on a control and user interface level - to web protocols. The moment you use web protocols it is so much easier to create user interfaces (e.g. using a Web browser as frontend) and it is simple integrate with existing systems (e.g. facebook). It would be interesting to assess how easy it is to use RESTful services to replicate some of the features of home automation systems. Sounds like an interesting project topic. There is a workshop on the Web of Things at PerCom in Mannheim - I am curious what is coming up there.

[1] James Scott, Boris Dragovic: Audio Location: Accurate Low-Cost Location Sensing. Pervasive Computing: Third International Conference, PERVASIVE 2005, Munich, Germany, May 8-13, 2005. Springer LNCS 3468/2005. pp 1-18. http://dx.doi.org/10.1007/11428572_1

No comments: