20.11.2013, 15.00h Talk by Philipp Scholl

Speaker: Philipp Scholl

Title: Activity Recognition with Instrumented Artifacts: The Wavelet Approach.

Abstract: The analysis of large time series remains a challenging task. With scientific projects like the Large Hadron Collider generating more than 41 terabytes of data a day. Even on a smaller scale, human activity recognition with body-worn sensors, or Wireless Sensor Network (WSN) in hard to control environments, a large effort needs to be put into the efficient lossy and lossless storage, fast retrieval of synopses and fast similarity search of the time series generated by the involved sensors. Activity recognition over the course of several weeks with body-worn kinetic sensors is an example where large time series are generated. Such data is typically recorded with high accuracy in long-term settings (like in-field studies), which involves sampling with frequencies above 50Hz. WSNs deployed over such long periods with a large amount of heterogenous sensors, used for example to monitor long-running experiments, are similarly challenging. Such long-term, high-fidelity monitoring with sensors embedded into physical artifacts are the application targets of this thesis while focusing on the fast retrieval of synopses and fast similarity search using the wavelet transform.