Probabilistic Inferencing for Location

INTRODUCTION Automatically measuring the location of a person or device for ubiquitous computing always involves the conversion of a raw measurement into a location measurement. While this process is a prepackaged part of some sensors (e.g. GPS and cell phones), researchers are often faced with making this conversion on their own as part of their efforts to deploy novel sensing technologies. This paper describes and discusses various general techniques that researchers have adopted for processing sensor readings into location measurements, emphasizing probabilistic approaches. Reasoning probabilistically is attractive because it naturally accounts for the uncertainty and ambiguity of sensor data, and a probabilistic representation is a good way to communicate uncertainty to higher level modules that exploit location. The paper concludes by advocating recursive filtering as the best general technique to use, with particle filtering having a slight advantage over the next best recursive technique, the hidden Markov model.