Minimizing Calibration Effort for an Indoor 802.11
Device Location Measurement System
John
Krumm and John Platt
Microsoft Research
Using an 802.11 wireless client as a location
sensor is an increasingly popular way of enabling location-based services.
Triangulation on signal strengths from multiple access points can be used to
pinpoint location down to a few meters. However, this level of accuracy comes
at the price of a manual, tedious, spatially high-density calibration of signal
strength as a function of location. This paper presents a new 802.11 location
algorithm based on a relatively coarse calibration. This helps answer the question
of how accurate location can be computed based on a realistic level of
calibration effort. The algorithm uses an interpolation function that gives
location as a function of signal strength. As such, it is suited to maintaining
some degree of performance in spite of reduced calibration data. We use this
feature to test the effect of reducing the number of calibration readings per
location and the number of locations visited during calibration. Our
experiments show that calibration effort can be significantly reduced with only
a minor reduction in spatial accuracy. This effectively diminishes one of the
most daunting practical barriers to wider adoption of this type of location
measurement technique.
Return to Machine Learning and User Interface workshop page.