In this paper we discuss the current situation of the activity recognition research field. In our opinion activity recognition approaches of today focus too much on algorithmic solutions for detecting whatever activities. The overall motivation or a usage scenario is often missing. Evaluations are often using an artificial data set and thus, many recognition approaches have not been tried in any real application context. Our suggestion is to unify publications and workflows to improve the overall quality and increase the reusability of knowledge even across other research domains. We propose a standard set of sections each publication has to include: motivation, design, evaluation, and lessons learned. Especially the sections motivation and evaluation should be particularly aware of the humans' needs. We think there is also a need for more different conferences or workshops focussing on one single application domain. Finally, researchers could start to accept activity recognition as a technology, enabling and supporting established or emerging fields like e.g. wellness informatics.
05 / 2010
Haptic, Audio and Visual Interfaces for Maps and Location Based Services