An Indoor Positioning Method Based on Regression Models with Compound Location Fingerprints

Abstract

An indoor positioning method is proposed and evaluated. The method combines location fingerprinting and dead reckoning differently from conventional combinations. It utilizes a compound location fingerprint, which is composed of radio fingerprints at multiple points of time, that is, at multiple positions, and displacements between them estimated by dead reckoning. To avoid accumulated errors from dead reckoning, the method uses short-range dead reckoning. The method was evaluated in a student room whose size was $11 \times 5 \mathrm{m}$ and had furniture. Six Bluetooth beacons were placed in the room. The received signal strength indicator (RSSI) values of the beacons were collected at 28 measuring points, which were points of intersection on a 1-m by 1-m grid where no obstacles existed. A compound location fingerprint is composed of RSSI vectors at two points and a displacement vector between them. A support vector machine (SVM) and random forests (RF) were used to build regression models. The root mean square error (RMSE) of position estimation with SVM and (RF) was respectively 2.35 and 0.80m. These errors were lower than those with a single-point baseline model, where a feature vector is composed of only RSSI values at one location. The results suggest that the proposed method is effective in indoor environments. We also discuss the relationship between the permissible error tolerance and cumulative percentages of correct answers, and the influence of dead reckoning errors on RMSE.