An Automated Indoor Localization System for Online Bluetooth Signal Strength Modeling Using Visual-Inertial SLAM

Abstract

Indoor localization is becoming increasingly important but is not yet widespread because installing the necessary infrastructure is often time-consuming and labor-intensive, which drives up the price. This paper presents an automated indoor localization system that combines all the necessary components to realize low-cost Bluetooth localization with the least data acquisition and network configuration overhead. The proposed system incorporates a sophisticated visual-inertial localization algorithm for a fully automated collection of Bluetooth signal strength data. A suitable collection of measurements can be quickly and easily performed, clearly defining which part of the space is not yet well covered by measurements. The obtained measurements, which can also be collected via the crowdsourcing approach, are used within a constrained nonlinear optimization algorithm. The latter is implemented on a smartphone and allows the online determination of the beacons' locations and the construction of path loss models, which are validated in real-time using the particle swarm localization algorithm. The proposed system represents an advanced innovation as the application user can quickly find out when there are enough data collected for the expected radiolocation accuracy. In this way, radiolocation becomes much less time-consuming and labor-intensive as the configuration time is reduced by more than half. The experiment results show that the proposed system achieves a good trade-off in terms of network setup complexity and localization accuracy. The developed system for automated data acquisition and online modeling on a smartphone has proved to be very useful, as it can significantly simplify and speed up the installation of the Bluetooth network, especially in wide-area facilities.

Keywords: Bluetooth low energy; beacon; constrained optimization; indoor localization; particle swarm optimization; path loss model; visual-inertial SLAM.

Figure 1

The ARCore tracking algorithm calculates transformation from camera C. S. to world C. S.

Figure 2

The concept of the ARCore concurrent odometry and mapping algorithm [55].

Figure 3

Locations of BLE beacons (green squares) and walked path (determined with the visual-inertial SLAM) during automated measurement capture. The ground truth path is marked with the red dashed line. The white pixels represent empty space and grey pixels represent occupied space.

Figure 4

Fitting the online constructed path loss models (2) to the measurements of the signal strengths after 30, 70 and 150 s of walking (for the BLE beacon with MAC address DA:57:30:EC:6A:D1). Since the data contain a lot of noise, the coefficient of determination for all models is quite low: $R^2=0.5$.

Figure 5

The upper part of the figure shows the error between the beacon’s actual location (with MAC address DA:57:30:EC:6A:D1) and the position calculated by the constrained nonlinear optimization method. The error depends on the time or the travelled distance. The lower part of the figure shows the BLE signal range (the difference between maximum and minimum signal strength) as a function of time or the travelled distance.

Figure 6

The indoor positioning results (after 5 min of walking) obtained using the online constructed path loss models and the PSO localization method.

Figure 7

The cumulative distribution functions for the positioning errors in determining the positions by the PSO method after 30, 70 and 310 s of walking.

Figure 8

The mean localization error as a function of time or the travelled distance. The decrease of error can be described by an exponential model $e=0.35exp(-0.04$t$)+0.98exp(-0.0006$t) with the coefficient of determination $R^2=0.8$.