Smart-DAYS

People

This project is co-ordinated by Dr Daniel Roggen.

Smart Distributed daily living ActivitY-recognition Systems

Smart-DAYS is motivated by the need to recognize complex human activities from body-worn sensors for assistive technologies, such as providing “memory reminders” to persons with memory loss, or supporting independent living.

Such applications are characterized by an open-ended environment and the set of relevant activities may not be known at design-time. In Smart-days we set to address this by novel data processing techniques combining:

• Unsupervised hierarchical clustering and semi-supervised learning to support high-level activity recognition
  
• Crowdsourcing to exploit the knowledge acquired by other users’s devices and web databases to bootstrap activity recognition. 

Smart-DAYS architecture

Smart-Days consists of wearable systems and a cloud back-end.

The wearable system comprises sensing and computing capabilities. It performs activity recognition for its user using activity models available from the cloud. It collects sporadic user activity annotations and uses self-learning to discover hierarchical sensor data structure corresponding to repeating hierarchical human activity patterns.

The cloud collects information relayed by the local devices and expands and restructures common shared activity models with incremental semi-supervised learning techniques. Additional web data are used to categorize user annotations,guide the hierarchical organization of the activity models, and aid in providing semantic meaning to repeating sensor data patterns.

ETHZ Wearable Computing Lab's team

The focus lies on the exploitation of crowd-sourcing to bootstrap activity recognition systems along two lines of work:

heig-vd iict's team

The focus lies on methods that to support the user's self-labeling of activities by making it more intuitive:

• Unsupervised timescale discovery
• Unsupervised hierarchical structure discovery
• High-level activity recognition and integrated methodology evaluation

Acknowledgements

Smart-DAYS is funded by the Hasler Foundation within the "SmartWorld" funding programme.

Publications