TR#540: Predicting Daily Behavior via Wearable Sensors
Brian Clarkson and Alex Pentland
Working Paper
July, 2001
We report on ongoing research into how to statistically represent
the experiences of a wearable computer user for the purposes of day-to-day behavior
prediction. We combine natural sensor modalities (camera, microphone, gyros) with
techniques for automatic labeling from sparsely labeled data. We have also taken
the next required step to build robust statistical models by beginning an extensive
data collection experiment, the "I Sensed" series, a 100 day data set consisting
of full surround video, audio, and orientation.
Project page: http://www.media.mit.edu/~clarkson/isensed
Keywords: contextual computing, peripheral sensing, Hidden Markov Models (HMM),
computer vision, computer audition, wearable computing
PDF . Full
list of tech reports