On the issue of variability in labels and sensor configurations in activity recognition systems
Daniel Roggen, Kilian Förster, Alberto Calatroni, Andreas Bulling, Gerhard Tröster
Proc. "How to do good activity recognition research? Experimental methodologies, evaluation metrics, and reproducibility issues" (Pervasive), pp. 1–4, 2010.
Abstract
Two aspects of the design and characterization of activity recognition systems are rarely elaborated in the literature. First, the influence of system performance with variability in sensor placement and orientation is often overlooked. This is important for the deployment of robust activity recognition systems. Second, the influence of labeling variability is also overlooked, especially w.r.t. label boundary jitter and labeling errors. This is important during the development of an activity recognition system as acquiring labels is costly. We argue that there is a need to explicitly address the consequences of such variability in publications, together with the mitigation strategies that are used. Elaborating on this is required to move the state of the art towards real-world applications, such as in industrial wearable assistance applications or pervasive healthcare.Links
Paper: roggen10_pervasive.pdf
BibTeX
@inproceedings{roggen10_pervasive,
author = {Roggen, Daniel and F{\"{o}}rster, Kilian and Calatroni, Alberto and Bulling, Andreas and Tr{\"{o}}ster, Gerhard},
title = {On the issue of variability in labels and sensor configurations in activity recognition systems},
booktitle = {Proc. "How to do good activity recognition research? Experimental methodologies, evaluation metrics, and reproducibility issues" (Pervasive)},
year = {2010},
pages = {1--4}
}
