Thursday, 12 April 2007

Wearable Activity Recognition – Talk by Kristof van Laerhoven


How many sensors do we need? That was one point in the discussion after Kristof’s talk. His approach, in contrast to many others, is to use a large number of sensors for activity recognition. This offers more freedom with regard to placement of sensors, variations of sensors, and also provides redundancy but makes the overall system more complex. His argument is that in the long term (when sensors will be an integral part of garments) the multi-sensor approach is superior – let’s wait some 10 years and then discuss it again ;-)

From a scientific perspective and in particular for machine learning (where he sees the greatest challenge) the larger number of distributed sensors is more interesting. I find his porcupine2 (http://porcupine2.sourceforge.net/) stuff quite interesting.

Kristof’s web page: www.mis.informatik.tu-darmstadt.de/People/kristof