—Lifelogging is a phenomenon whereby an individual
digitally records his/her personal life experiences, for a variety of
purposes. Activity recognition and segmentation is fundamental
to many of the use cases in lifelogging. However, detecting
sufficiently robust user activity boundaries that could be deployed
with confidence in a subjective real-world setting remains a
challenge. In this paper, we extend our previous work on identifying a better activity recognition and segmentation approach
to multimodal lifelog data, primarily through the introduction of
automatic thresholding techniques, but also through revising the
criteria for selecting the most appropriate size of sliding window
when evaluating the proposed algorithms. We use an open and
publicly available lifelog test collection over a time period of
27 days with manual annotations and manually groundtruthed
activities.
Metadata
Item Type:
Conference or Workshop Item (Paper)
Event Type:
Conference
Refereed:
Yes
Uncontrolled Keywords:
Image segmentation; Metadata; Visualization; Activity recognition;
Manuals; Indexing; TV