Generic performance metrics for continuous activity recognition

Authors:
Albert Hein;Thomas Kirste
Affiliations:
Dept. of Computer Science, University of Rostock, Germany;Dept. of Computer Science, University of Rostock, Germany
Venue:
KI'11 Proceedings of the 34th Annual German conference on Advances in artificial intelligence
Year:
2011

Citing 3
Cited 0

Information Retrieval

Information Retrieval
Performance metrics for activity recognition

ACM Transactions on Intelligent Systems and Technology (TIST)
Evaluating performance in continuous context recognition using event-driven error characterisation

LoCA'06 Proceedings of the Second international conference on Location- and Context-Awareness

Quantified Score

Hi-index	0.00

Visualization

Abstract

For evaluating activity recognition results still classical error metrics like Accuracy, Precision, and Recall are being used. They are well understood and widely accepted but entail fundamental problems: They can not handle fuzzy event boundaries, or parallel activities, and they over-emphasize decision boundaries. We introduce more generic performance metrics as replacement, allowing for soft classification and annotation while being backward compatible. We argue that they can increase the expressiveness and still allow more sophisticated methods like event and segment analysis.