Pervasive - Tutorial: John Krumm: Processing Sequential Sensor Data
Abstract:Sequential sensors are those that product a sequence of sensor readings of the same entity over time, such as GPS and accelerometers. Measurements from sensors like these are important for pervasive computing, because they are used to infer a person's context. Unfortunately, sensors are never perfect in terms of noise or accuracy, and they often do not measure the state variables we really need. This tutorial is aimed at introducing fundamental techniques for processing sequential sensor data to reduce noise and infer context beyond what the sensor actually measures. The techniques discussed are not necessarily on the cutting edge of signal processing, but they are well-accepted approaches that have proven to be fundamentally useful in pervasive computing research. Specifically, this tutorial discusses mean and median filters, the Kalman filter, the particle filter, and the hidden Markov model (HMM). Each of these techniques processes sequential sensor data, but they all have different assumptions and representations, which are highlighted to help you make an intelligent choice. The techniques will be illustrated with a running example.