Discovery of recurring patterns from sensor data (Current Research)
Sensors may include on-body accelerometers, microphones, and video cameras in the environment
Learn models of common primitive actions (short, stable movements similar to movemes in existing literature) associated with particular activities
Aggregate primitive actions to learn higher level models for prediction and recognition
Recognizing Activity from Video
Detect primitive events and actions
Recognize compound and extended activities
Represent the temporal and causal structure of an activity
Learn event patterns to characterize actions
Adapting low-level vision processing based on high-level feedback
Build activity monitoring systems that leverage active interaction with users
Example: integrate information from a vision-based tracking system in an office with recognized activities that imply location (opening a door, typing, sitting, etc.)
Activity Mining in Sensor Networks
C.R. Wren and D. Minnen Advances in Neural Information Processing Systems (NIPS)
Workshop on Activity Recognition and Discovery, December 2004.
Activity Discovery Overview
Problem introduction and overview of activity discovery algorithm (March 1, 2006)
Expectation Grammars Overview
Overview of activity recognition research using stochastic grammars (presented as poster at CVPR, June 2003; slides from qualifier orals, Dec 2003)
Modeling Human Activities
Exploring how to learn scripts and bridge a perceptual/semantic reasoning gap
Cognitive Science Colloquium, October 17, 2003