MIND: A Nonparametric Decision Fusion Method for Accurate Indoor Localization using Sensors with Monotonically Increasing Error Functions

A nonparametric fusion method for extracting accurate distance measurements from low-quality sensors is proposed. The method applies to sensors with error functions that are monotonically increasing with respect to (w.r.t.) the actual value to be measured (arguments are presented on why a monotonically increasing error function is something to be expected with range-estimating sensors). The proposed method has been developed in order to enhance the performance of localization systems that utilize commercially available sensors for range estimation to achieve localization through triangulation of range estimates. The proposed method is based on evaluating multiple sensor measurements and using the minimum measured distance as a more efficient estimate of the real distance compared with calculating and selecting the distance average. Thus, the proposed method is code-named MIND (from MINimum Distance). It is shown analytically that MIND outperforms, in terms of location estimation accuracy, the sensor with the minimum mean error when used in a multi-sensor configuration. An experimental testbed consisting of four Cricket sensors in a symmetric bundle configuration was used to evaluate the MIND fusion method experimentally. For each Cricket sensor, performance characteristics were established through extensive laboratory analysis and were found to yield highly inaccurate range estimates. However, when these low-quality Cricket sensors were fused in a four-sensor symmetric configuration, it was shown experimentally that the MIND fusion method exhibits near-optimal performance and largely overcomes most of the flaws of the underlying low-quality Cricket sensors, delivering a localization solution of extended accuracy, availability, and robustness.