Please send your 3 slides for the one-minute-madness a.s.a.p. to email@example.com. Thanks.
Activity recognition (AR) emerged as a key area of research in pervasive computing and plays a central role in the field’s vision of developing context-aware systems and interaction. AR has been applied to a wide range of domains such as healthcare, well-being, quality management, and human-computer interaction. The general objective for current activity recognition systems is to analyze continuous sensor data streams and to detect and discriminate certain activities of interest, and to reject portions of sensor data that cover unknown or idle activities. This corresponds to a classical, open lexicon pattern classification task. The majority of AR applications currently address (variants of) this general classification task. Rare exceptions comprise, for example, activity spotting – the problem of identifying the begin and end of an activity in the continuous data stream. A wealth of algorithmic approaches to activity recognition have been developed that employ state-of-the-art signal processing and machine learning techniques.
Reviewing the current literature in the field may give the impression that the general problem of activity recognition has almost being solved. State-of-the-art AR systems achieve recognition rates of well beyond 80 or 90 percent across tasks. This opens up the question which other – more challenging – activity recognition problems should be addressed in the future. More generally speaking, this also relates to the question how a future AR system should look like. This is in concordance to the results of a related workshop at the 2010 Pervasive conference. In that workshop the formulation of “grand challenges” for future activity recognition tasks has been discussed. These challenges go beyond the aforementioned classification approach and could, for example, comprise a quan- titative measure of the quality of performing an activity rather than only detecting an activity. The well-explored example of analysis of activities of daily living (ADL) would then extend to monitoring how well these activities are being pursued. A number of practical applications are imaginable, such as comparisons between persons (e.g. for sports-related tasks), long-term monitoring of single person’s behaviors (e.g. for therapeutical applications) or the detection of abnormal / novel activity patterns.
It is questionable whether commonly used techniques, such as sliding-window approaches and supervised classifier training, are sufficient to achieve this goal. Instead, addressing these grand challenges will likely require the AR community to develop new activity recognition chains (ARC) that include techniques for signal processing, feature extraction, classification, and evaluation. Furthermore, related benchmark datasets would require enhanced ground truth annotations that cover the activities to be analyzed in much more detail.
IWFAR 2011, a satellite workshop of Pervasive 2011, is supposed to focus on new “frontiers in activity recognition using pervasive sensing” in the aforementioned sense. We want to stimulate and explore the creativity of the community regarding new applications and approaches to AR. The latter could comprise radically new procedures like, for example, biologically inspired AR methods or rigorously exploiting general time-series analysis approaches (e.g. from the financial domain). Recent publications in the AR community represent quite promising starting points (cf. e.g. time-delay embedding for the analysis of repetitive activity patterns, which was inspired by the physics of complex systems).
IWFAR 2011 provides a forum for researchers and practitioners to gather and present new ideas, and discuss aspects related to applications and techniques that go beyond classic activity recognition. Covering a more or less “emerging” field, expected contributions would also cover more speculative ideas that are too specific for the broader audience of the main conference. We solicit high-quality technical papers that shall be presented orally while leaving enough time for discussions.